fMRI correlates of retrieval orientation

Abstract

Processes related to the cueing of memory retrieval were investigated using fMRI during a yes/no recognition memory test. Participants studied either pictures or auditory words and were subsequently tested with visual words that either corresponded to studied items (old items) or were unstudied (new items). It was expected that neural activity would differ according to the type of information participants are trying to retrieve: a manifestation of the so-called “retrieval orientation” effect. We replicated robust old/new effects in parietal, prefrontal and anterior medial temporal lobe cortices, which did not differ according to the study material. However, we did observe differential activity to test items in temporoparietal cortex and fusiform cortex as a function of the study material. More specifically, attempts to retrieve words encoded auditorily produced greater activity in auditory cortex than attempts to retrieve words encoded as pictures, whereas the converse was found in fusiform cortex. The above pattern was found for both new and old test items. These findings implicate these regions in constraining the search for specific types of encoded information and thus are in accordance with the transfer-appropriate processing framework. Further, we propose that our results can be seen as an extension of the cortical reinstatement hypothesis, in that the same material-specific cortical regions are engaged during both encoding and retrieval, and this increases the likelihood of successful recognition, or rejection, of retrieval cues in a memory test.

Introduction

Since the time of William James (1890), memory researchers have been concerned with the interactions between retrieval cues and stored memory representations. In 1983, Tulving proposed that two conditions must be met to evoke these interactions. First, one must be in a certain cognitive state called “retrieval mode”, and second, a cue must be effective in triggering memory retrieval (Tulving, 1983). The effectiveness and importance of retrieval cues had already been recognized in the “transfer-appropriate processing” principle of Morris et al. (1977) and in the “encoding specificity” principle of Tulving and Thompson (1973). In general terms, these principles state that successful retrieval is a function of the degree of overlap between the processes engaged at encoding and those engaged at retrieval.

Rugg and Wilding (2000) proposed that in order to increase the likelihood of successful retrieval, people can vary how they process retrieval cues so as to align cue processing with the demands of a particular retrieval goal. They called these cognitive states “retrieval orientations”, which were conceived as a fractionation of retrieval mode, although the precise relationship between retrieval orientation and retrieval mode is yet to be established (but see Herron & Wilding, 2004). Rugg and Wilding proposed that, in contrast to the neural correlates of retrieval mode, which can be observed by comparing different classes of retrieval task (episodic versus non-episodic), the neural correlates of different retrieval orientations can be investigated by comparing situations where the same class of retrieval cue is used to search memory for different kinds of information, while holding all other aspects of the retrieval task constant.

Several event-related potential (ERP) studies have used this approach within the yes/no recognition memory paradigm. The neural correlates of retrieval orientation have been isolated by either examining cue-related activity preceding the actual test item (Herron & Wilding, 2004), or by restricting analysis to unstudied items (Herron & Rugg, 2003; Hornberger et al., 2004; Hornberger, Rugg, & Henson, in press; Robb & Rugg, 2002; Rugg et al., 2000). In the latter case, the use of unstudied items minimizes contamination of the ERP by any episodic retrieval that might occur as a consequence of cueing memory (i.e. “retrieval success”). When comparing attempts to retrieve items studied as pictures with attempts to retrieve items studied as words, the ERP retrieval orientation effect typically starts around 250 ms after onset of an unstudied item, and is often sustained until the end of the recording epoch (Hornberger et al., 2004, submitted for publication; Robb & Rugg, 2002).

The inability to adopt an appropriate retrieval orientation, i.e. to process retrieval cues in a manner that increases the probability of successful retrieval, may underlie some of the memory problems associated with healthy ageing (Jacoby, Shimizu, Velanova, & Rhodes, 2005). This is consistent with the reduced amplitude of the ERP retrieval orientation effect in older people (Morcom & Rugg, 2004). Whether the amnesia associated with various types of brain damage can include an impairment of retrieval orientation is yet to be investigated fully.

The ERP retrieval orientation effects reported in the above studies are generally widespread over the scalp, with an amplitude maximum at mid-central electrodes when using a mastoid reference (though see Dzulkifli et al., 2004; Herron & Wilding, 2004; Johnson et al., 1997, Rugg et al., 2000, for more lateralized, frontal scalp distributions). This widespread activity makes it difficult to estimate the neural generator(s) of the ERP effect, particularly given that the brain and skull act as spatial low-pass filters for electrical fields.

One potential solution to this problem of identifying the neural generators of retrieval orientation effects is to use haemodynamic methods like fMRI. To our knowledge, only a handful of fMRI studies have contrasted conditions that would meet the above criteria for isolating retrieval orientation.

Ranganath et al. (2000) (based on a previous ERP experiment, Ranganath & Paller, 1999), contrasted recognized and rejected items according to different levels of specificity of retrieval. In the study phase, participants made size judgements about visually presented objects. In a subsequent test phase, they judged whether objects had been presented in the same size at study (their “specific” condition) or whether objects had been seen at all, regardless of their size (their “general” condition). Ranganath and colleagues found that regardless of the study status (old/new) of the test items, left anterior prefrontal cortex (BA 10/46) showed greater activity in the specific than general conditions. Thus, left prefrontal cortex might be viewed as a likely candidate for a neural origin of retrieval orientation, being more active when participants need to focus their search of memory for specific information, regardless of whether that search is successful. However, a subsequent study (Rugg et al., 2003) failed to activate left prefrontal cortex when comparing a condition that required retrieval of specific source information with a condition that did not. In fact, Rugg et al. found greater activity in the left prefrontal region identified by Ranganath et al. for correct recognition of old items than for Correct Rejection of new items, suggesting that this region is sensitive to retrieval success. The reason for the difference between the two sets of findings is unclear, but could relate to the different material used (pictures versus words), the different type of source information tested (size versus position/color) or differences in the precise task instructions (e.g. the use of three responses in the Specific condition of Ranganath et al., but only two in their General condition; see Rugg et al., 2003, for further discussion).

Another fMRI study by Dobbins et al. (2003) contrasted neural activity in a source task with activity in a recency task for both successful and unsuccessful retrieval. At test, participants were presented with two words, and cued to indicate either (1) which one was encoded in the context of a specific study question or (2) which one was studied more recently, regardless of study question. They proposed that retrieval orientation would differ for the two types of test condition. Like Ranganath et al. (2000), Dobbins et al. found that lateral prefrontal (and lateral parietal) regions were more active in the source than recency task, regardless of whether or not the two-alternative choice was correct. However, the use of two old items at test (rather than a single item that was either old or new, as in the Ranganath et al., 2000, Rugg et al., 2003, studies) complicates the comparison of the “retrieval success” effects across the studies (as does the use of yet another type of source information).

One potential problem with all three of the above fMRI studies is that the contrasts used to isolate retrieval orientation may be confounded by generic difficulty effects (or “retrieval effort”, Buckner et al., 1998), given that accuracy tended to be lower, and/or reaction times longer, for the source memory tasks. While Rugg et al. (2003) factorially varied retention interval, and Dobbins et al. (2003) varied inter-item lag, in attempts to address difficulty effects, neither manipulation was completely successful. More fundamentally, all three studies tested retrieval orientation by changing the retrieval task. According to Rugg and Wilding's (2000) proposal, a manipulation of the type of study material being sought, while holding the retrieval task constant, should also induce differences in retrieval orientation. The recent ERP data using this approach, mentioned earlier, have shown that the retrieval orientation effect associated with different types of study material is not affected by factors that affect difficulty (but see Dzulkifli & Wilding, 2005). For example, in Experiment 2 of Hornberger et al. (2004), there was an effect of retrieval orientation on new item ERPs even though accuracy and reaction time did not differ reliably across the conditions.

Thus, in the present study we used the same design as in Experiment 2 of Hornberger et al. (2004): in one type of block, participants studied pictures and were then tested for yes/no recognition using visual words (half of which were the names of studied pictures); in the other type of block, participants heard words at study before being tested for yes/no recognition using visual words (half of which again corresponded to studied items; see Fig. 1). By contrasting correctly rejected new test items as a function of whether participants were trying to remember pictures or auditory words, we hoped to identify regions sensitive to retrieval orientation.