Cerebral Cortex Advance Access originally published online on December 28, 2005
Cerebral Cortex 2006 16(11):1614-1622; doi:10.1093/cercor/bhj098
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Isolating Rule- versus Evidence-Based Prefrontal Activity during Episodic and Lexical Discrimination: a Functional Magnetic Resonance Imaging Investigation of Detection Theory Distinctions
Department of Psychological and Brain Sciences, Duke University, Durham, NC 27708, USA
Address correspondence to Ian G. Dobbins, PhD, Psychological & Brain Sciences, Duke University, PO Box 90086, 9 Flowers Drive, Durham, NC 27708, USA. Email: ian.dobbins{at}duke.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and...ResultsDiscussionReferences |
|---|
Dorsolateral and frontopolar prefrontal cortices (PFCs) are often implicated in neuroimaging studies of memory retrieval, with this activity ascribed to controlled monitoring processes indicative of difficult or demanding retrieval. Difficulty, however, is multiply determined, with success rates governed both by the available evidence and by the nature of decision rules applied to that evidence. Using event-related functional magnetic resonance imaging, we isolated these factors by 1) contrasting different decision rules across matched evidence and 2) manipulating the level of evidence within a fixed decision rule. For identically constructed retrieval probes (1 old and 1 new item), same–different (are these different?) compared with forced–choice (which one is old?) decision rules yielded bilateral dorsolateral and right frontopolar PFC increases. However, these regions were unaffected when the available evidence was greatly lowered within forced–choice decisions. Thus, the regions were simultaneously sensitive to the type of decision rule and yet insensitive to the level of evidence supporting those decisions. Analogous lexical tasks yielded similar patterns, demonstrating that the PFC responses were not episodic memory specific. We discuss the mechanistic differences between same–different versus forced–choice decisions and the implications of these data for current theories of PFC activity during episodic remembering and executive control.
Key Words: executive control • fMRI • memory • prefrontal cortex • signal detection theory
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and...ResultsDiscussionReferences |
|---|
One of the more surprising findings arising from the application of functional imaging to episodic memory retrieval has been the frequent observation of prefrontal cortex (PFC) activity across a wide range of memory judgments (Nyberg and others 1996
; Cabeza and others 1997; Duzel and others 1999). Many characterizations of this activity rely upon the notion that PFC responses are indicators of the quality or quantity of mnemonic evidence available during the memory demand. For example, dorsolateral PFC has been suggested to reflect controlled retrieval monitoring operations in which the output of episodic memory is evaluated with respect to a the currently held retrieval goal or criterion (Nolde and others 1998; Henson and others 2000; Ranganath and others 2000; Dobbins and others 2002; Rugg and others 2003), and the degree of monitoring is often held to reflect the availability of supporting memory evidence. That is, it is assumed that greater retrieval monitoring is required as memory evidence becomes increasingly impoverished, scarce, or ambiguous with respect to the desired criterion. For example, in a comparison of response confidence during recognition, Henson and others (2000) found that low- versus high-confidence decisions were accompanied both by declining performance and greater activity in right dorsolateral PFC. This led to the conclusion that as memory evidence approached the observer's cutpoint or criterion for classification, greater monitoring was required and hence activity increased in the dorsolateral PFC region (see also Henson and others 1999; Rugg and others 2003). Somewhat similarly, Velanova and others (2003) examined both blocked and trial-by-trial activity levels during recognition retrieval of well versus more poorly encoded verbal material and found that responses in dorsal/midventral and frontopolar regions were elevated during attempted retrieval of poorly versus well-encoded material in both a blocked (frontopolar) and event-related (dorsolateral/midventrolateral) fashion. The elevated responses were assumed to reflect the need for greater executive control processes during retrieval of impoverished versus well-encoded memory content.
Thus, at present, monitoring and executive control characterizations suggest that PFC responses during episodic retrieval are a necessary consequence of increasingly impoverished or insufficient memory evidence, and in this sense, these activations reflect retrieval difficulty. However, these characterizations are potentially incomplete because task difficulty (as indexed by raw success rates) is typically modeled as the result of at least 2 independent causes. For example, signal detection theory (SDT) assumes that 2 independent factors govern observed success rates during item discrimination tasks (Macmillan and Creelman 1991). The first, as discussed above, is the quality of the mnemonic evidence. For a fixed discrimination task, subjects will be more successful given more extensive memory encoding, and this represents an "evidence-based" modulation of performance. However, the other major factor moderating success rates is the nature of the decision rule that observers choose or are directed to apply to the available memory evidence. Even when the evidence is fully matched across conditions, the use of different decision rules can result in very different success rates, and this represents a "rule-based" modulation of performance. Thus, under SDT, changes in the decision rules and/or changes in the strength of evidence jointly contribute to observed success rates but do so via functionally independent mechanisms. If correct, evidence- versus rule-based modulations of performance should implicate different cortical regions during discrimination.
To test this model, we contrasted forced–choice and same–different decision rules, which under SDT predict very different success rates even when applied to probes containing equivalent evidence (Fig. 1). During forced–choice, subjects attempt to select a target from among N lures, and in the simplest case, "2-alternative forced–choice" (2AFC), the observer is presented a target and 1 lure. Under SDT, such judgments involve a relative comparison of evidence across the items; the item with the most evidence is classified as the target. In contrast to forced–choice judgments, same–different judgments are assumed to require fundamentally different decision operations. Here the observer must determine whether the items originate from the "same" or "different" classes, and this requires the observer to instead independently classify each member of the pair in order to conclude whether they are similarly constructed (same) or are from different classes (different) (for review, see Macmillan and Creelman 1991). What is critical about these decision rules is that they can both be applied to identically constructed retrieval probes. That is, observers can be given forced–choice memory discriminations, during which all probes contain an old and a new item, and also given same–different memory discriminations, during which a large and critical portion of the trials contain probes containing an old and a new item and for which the correct response is different. (Trials in which the items are actually the same [i.e., both old or both new] are also used during same–different judgments but are of limited value during imaging because they differ both in content and in rule from forced–choice decision trials.) Under SDT, although the probes are identical across these 2 trial types, the requirement to employ different decision operations will result in the proportion of correct responses during same–different judgments (different) being considerably lower than the proportion of correct forced–choice judgments (Fig. 1). This constitutes a purely rule-based modulation of success rates because the probes in these 2 cases are equivalent.
|
Here, we use this manipulation to test the hypothesis that recruitment of PFC regions during episodic retrieval is tied to rule- and not evidence-based modulations of performance. We tested this during episodic recognition by contrasting 2AFC and "2-alternative same–different" (2ASD) trials for identically constructed retrieval pairs containing 1 old and 1 new item. Furthermore, we verified that observed PFC differences reflected the use of these different decisions rules and were not somehow the consequence of the theoretically predicted differential success rates. To do this, we modulated performance within the forced–choice task by including a 3-alternative forced–choice (3AFC) condition during which an additional lure was added. During "3AFC," the level supporting evidence for judgments is impoverished relative to 2AFC, and thus, success rates will necessarily fall. If PFC responses were strictly tied to the level of supporting evidence and increased when such evidence became impoverished, then 3AFC should yield greater activity than 2AFC. In contrast, if activity differences between same–different and forced–choice conditions reflected a fundamental difference in the nature of the recruited decision operations, then forced–choice activity should not change with the addition of a lure.
Finally, we tested the specificity of any observed rule-based modulations of PFC to the episodic memory domain. The SDT account applies generally to discriminations involving continuous and noisy evidence, and therefore, similar regions may be implicated across episodic and nonepisodic domains, provided the nature of the decision rules is maintained (cf., Duncan 2001). Furthermore, previous functional imaging research suggests that several PFC regions may play a common role across a range of cognitive tasks, leading to the idea that certain PFC regions participate in "cross-domain" cognitive functions (e.g., Cabeza and others 2003; Nyberg and others 2003). To test this, we changed the episodic recognition task to a lexical judgment on pairs containing the actual words "OLD" and "NEW." For forced–choice pairs, subjects indicated which printed word spelled OLD. For same–different trials they indicated whether the pairs were the same (e.g., "OLD OLD") or different words (e.g., "NEW OLD"). Despite the fact that these lexical discrimination tasks are patently easy, in principle they nonetheless involve qualitatively different decision operations, and thus, differential activity will potentially occur in the same PFC regions sensitive to the decision type during the episodic discriminations.
|
Materials and Methods/Experimental Procedures |
|---|
|
TopAbstractIntroductionMaterials and...ResultsDiscussionReferences |
|---|
Subjects
Twelve (19- to 28-year old and 6 females) native English-speaking volunteers were included in the study. Informed consent was obtained in a manner approved by the Institutional Review Board of Duke University Medical Center.
Materials
A total of 624 nouns were drawn randomly from a pool of 1216 words for each subject. From the list, 3 lists of 208 items (96 old and 112 new items for each cycle) were constructed for use in 3 study/test cycles. The items in the pool were, on average, 7.09 letters and 2.34 syllables, with a Kucera–Francis corpus frequency of 8.85.
Study Procedure
Stimuli were back projected onto a screen at the rear of the magnet bore and were viewed with a mirror placed above the eyes. All responses were made with the left hand using an optical button box with 4 available keys. During the study phase, subjects studied the list of items while counting the syllables of each word. This incidental encoding task has been found to yield moderate levels of recognition performance suitable for the current study (e.g., Dobbins and others 2000). The study phase had 144 trials including 36 fixation baseline trials interspersed among the counting syllable trials. During fixation trials, a cross was presented underneath the instruction "relax." Subjects were given 2 s to respond using 1 of the 4 keys to indicate the number of syllables. Half of the study items (48 out of 96 words) were presented twice to manipulate encoding frequency (once vs. twice). (The manipulation of frequency failed to significantly modulate same–different performance and provided only a modest change in forced–choice performance. Because this manipulation was conceptually redundant with the comparison of 2AFC and 3AFC [viz., a change in evidence], we simply collapsed across encoding frequency during test phase analyses to increase power.)
A test phase immediately followed each study phase. Subjects were presented with a word pair or a triplet in each retrieval trial for 6 s. During the 2AFC trials, subjects were presented with a word pair (1 target and 1 lure), underneath which the task question "which old?" was displayed indicating that they should select the studied word using the index or middle finger to indicate either the left or the right word, respectively. We also used a 3AFC recognition task (1 target, 2 lures). In contrast to these forced–choice trials, during 2ASD trials, subjects were required to determine whether the pair contained 1 old and 1 new item (different) or were constructed from the same item types (both studied or both novel—same). On half of the 2ASD trials, the probes were constructed identically as those used during 2AFC, and hence, the episodic evidence was matched across these conditions. For completeness, and to guard against response biases, pairs of the same items (either both old or both new) were included in the remaining half of the 2ASD trials; however, these are not analyzed because they differ in both content and rule from the critical 2AFC trials. The task prompt for 2ASD trials read "different?", and the index and middle fingers were used to indicate either a "yes" or a "no" response. The screen location of old and new items in these conditions was randomly assigned across trials. The order of 96 retrieval trials along with 32 fixation trials (25% of total trials) was determined using an optimal sequencing program to facilitate recovery of the blood oxygen level dependent signal (Wager and Nichols 2003).
During the lexical decision scan, subjects were presented with pairs containing the actual words OLD and NEW on the screen and again asked to make either forced–choice (which old?) or same–different (different?) classifications. All other aspects closely matched the forced–choice and same–different memory scans. There were 3 scans/runs of the episodic memory task and 1 scan/run of the lexical decision task (20 trials for each decision condition in addition to fixation/null trials). All subjects started with the lexical decision task.
Functional Magnetic Resonance Imaging Acquisition
Scanning was performed on a 1.5-T General Electric scanner using a standard head coil. Functional data were acquired using a standard spiral pulse sequence (repetition time = 2000, echo time = 40 ms, 24 axial slices parallel to anterior commissure–posterior commissure plane, 3.75 x 3.75 x 5.0 mm, no gap). Before functional data collection, 4 dummy volumes were discarded to allow for T1 equilibration. Participants' head motion was minimized using foam padding and a forehead strap. High-resolution three-dimensional full-brain spoiled gradient-recalled images were acquired to aid in visualization and normalization.
Functional Magnetic Resonance Imaging Analyses
Our analyses focused on the memory retrieval and lexical decision scans. Data were processed using statistical parametric mapping software (SPM99) (http://www.fil.ion.ucl.ac.uk/spm/). Slice acquisition timing was corrected by resampling all slices in time relative to the middle slice collected, followed by rigid body motion correction across all runs. Functional data were spatially normalized to a canonical echo planar imaging (EPI) template using a 12-parameter affine and nonlinear cosine transformation and then spatially smoothed with an 8-mm full-width at half-maximum isotropic Gaussian kernel. Each scanning session was rescaled such that the mean signal was 100.
The data were statistically analyzed, treating subjects as a random effect. For the analyses, volumes were treated as a temporally correlated time series and modeled by convolving a synthetic hemodynamic response function (HRF) and its first-order time and dispersion derivative using the onset times for the events.
The resulting functions were used as covariates in a general linear model, along with a basis set of cosine functions that were used to high-pass filter the data and a covariate representing session effects. The least squares parameter estimates of the best-fitting synthetic HRF for each condition of interest (averaged across scans) were used in pairwise contrasts and stored as a separate image for each subject. These difference images were then tested against the null hypothesis of no difference between contrast conditions using one-tailed t tests.
To control for familywise error resulting from multiple comparisons, a Monte Carlo simulation procedure was conducted to determine a height and extent threshold sufficient for enforcing a maximum type-1 error of 0.05 for the reported clusters (Slotnick and others 2003; Slotnick and Schacter 2004). The procedure iteratively simulated (1000 simulations) spatially smoothed noise for each voxel in the volume using unit normal random sampling and, for a given height threshold, estimated the likelihood of spuriously detecting clusters of different sizes. Based on the results of the simulation, clusters were considered significant and further scrutinized if they consisted of 29 or more contiguous voxels (3 mm isotropic) exceeding a minimum probability threshold of 0.003. Thus, the reported maps for all simple contrasts are cluster corrected for multiple comparisons at the 0.05 alpha level. For the conjunction of memory and lexical decision maps, inclusive masking was used to identify regions differentially active (2ASD > 2AFC) during both memory and lexical decision contrasts, which were drawn from separate scans. Areas demonstrating voxel effects at 0.01 for each of the 2 separate contrasts, with an extent of 29 or more overlapping voxels, were subjected to further analysis. Given that the contrasts were drawn from independent scans, this resulted in a final map with clusters of 29 or more voxels and a minimum joint probability of 0.0001 for each included voxel. Based on the Monte Carlo results noted above, this procedure is sufficiently conservative to ensure that reported clusters for the map were corrected at the 0.05 or better level.
Functional regions of interest were extracted using peristimulus time averaging for the event-related functional magnetic resonance imaging data surviving the above retrieval contrasts. Percent signal averages were obtained for the above threshold voxels within and 8-mm radius of each of the SPM-identified maxima. These extracted data were further analyzed using off-line statistical software.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and...ResultsDiscussionReferences |
|---|
Behavioral Data
As expected, subjects were less successful on the same–different compared with forced–choice trials containing matched evidence (2AFC vs. 2ASD, Fig. 1c). For trials in which the pair contained an old and a new item, subjects selected the correct item 77% of the time during forced–choice, whereas they correctly concluded that the pair was different on 53% of the trials (t(11) = 7.16, P < 0.001). The subjects correctly concluded that the pair was the same 61% of the time during same–different trials in which both items were old or both were new (not plotted). Adding a lure to the forced–choice task (3AFC) significantly lowered forced–choice success rates (M = 0.60, t(11) = 5.47, P < 0.001) to a level on par with same–different judgments (t(11) = 1.64, not significant [NS]). Overall, these success rates are consistent with what would be expected given an overall d' of approximately 1 across the different task variants.
With respect to median response times during correct trials, subjects were significantly slower during 2ASD (3000 ms) than 2AFC (2349 ms; t(11) = 10.28, P < 0.001). Similar to the effect on success rates, adding a lure to the forced–choice (2833 ms) significantly slowed forced–choice reaction time (t(11) = 7.03, P < 0.001), bringing it down toward levels observed during 2ASD judgments, although the latter remained reliably slower (t(11) = 2.20, P = 0.05).
Functional Imaging Data
A direct contrast of 2ASD versus 2AFC trials demonstrated significant increases in bilateral dorsolateral (approximate Brodmann's area [BA] 46/9), right frontopolar (
BA 10), and medial superior frontal (BA 6/8/32) PFC regions, in addition to increases in bilateral extrastriate areas (BA 18/19) (Fig. 2a and Table 1). No relative increases were observed during the reverse contrast. Thus, despite the fact that these trials contained memory probes that were identically constructed, and hence matched for available evidence, there were prominent differences in the PFC responses. This, in and of itself, demonstrates that these PFC responses are not direct indicators of the quality or degree of memory content elicited by the probes. However, it could be argued that the responses were somehow a secondary consequence, as opposed to the primary cause, of the performance differences across same–different compared with forced–choice judgment. To rule this out, we further contrasted 2AFC with a 3AFC condition in which an additional lure was added to the retrieval probes. In this contrast, the decision rule is matched, yet the supporting evidence differs. If simple performance differences or differences in the strength of supporting evidence were somehow causing the PFC activation differences, then a similar pattern should emerge when comparing 3AFC with 2AFC; that is, the lowered success rate and slower responding during 3AFC should be associated with increases in PFC regions similar to those implicated in the comparison of 2ASD to 2AFC. However, despite the fact that forced–choice performance was considerably worsened with the addition of a lure, a direct comparison of 3AFC with 2AFC revealed no increases in the aforementioned PFC regions (Fig. 2b). When the threshold was relaxed considerably (0.01, 29 voxel extent), differential activity still failed to emerge in dorsolateral and anterior PFC regions; however, dorsal-medial PFC (BA 6, Montreal Neurological Institute [MNI] [–6, 9, 57]) and a very posterior portion of the left PFC (BA 6/9, MNI [–42, 3, 33]) demonstrated slightly greater activity during 3AFC, consistent with the likelihood of slightly elevated verbal working memory requirements during the condition with more items (Cabeza and Nyberg 2000). Critically, however, neither the dorsolateral nor the right frontopolar regions were implicated. This null finding is consistent with the reconstructed time courses for these more anterior, rule-sensitive PFC regions, which demonstrated almost identical hemodynamic responses for forced–choice decision making regardless of the number of lures (Fig. 2; line plots 1–3).
|
|
Finally, for completeness, we directly contrasted 2ASD and 3AFC in order to formally verify that the greater activity remained for the same–different rule even when the reference task (3AFC) contained more items and yielded similar performance characteristics. Note that this analysis only identified regions that yielded significant activations that were also previously implicated in the initial 2ASD versus 2AFC contrast. Any additional regions were excluded from the map because their interpretation is ambiguous, given that the comparison confounds the nature of the decision rule (same–different vs. forced–choice) with the number of items present during the trial (2 vs. 3). As seen in Figure 2c, this contrast implicated all the PFC regions implicated in the 2ASD versus 2AFC contrast and therefore strengthens the conclusion that these regions are indeed sensitive to the decision rule and not strictly tied to the differences in relative performance.
In contrast to the pattern of dorsolateral and frontopolar PFC responses that reliably differed across the rules regardless of performance characteristics, the posterior visual association areas demonstrated greater activity either when the number of inspected items increased (Fig. 2b; line plot 4) or when the number of items was matched, but the rule changed (Fig. 2a; line plot 4). This constitutes a notable dissociation between the response of extrastriate and the dorsolateral and anterior PFC regions, with the former likely sensitive to the time spent visually attending to the probe display regardless of task rules and the latter sensitive to the nature of the recruited decision operations regardless of time on task.
The above contrasts implicated bilateral dorsolateral and right frontopolar regions in decision operations that are increasingly necessary for, or fully isolated to, the same–different as opposed to the forced–choice memory judgments. To determine whether these operations were unique to episodic memory retrieval, we contrasted the same decision rules during a simple lexical decision task in which subjects were presented with combinations of the actual word stimuli OLD and NEW and required to either select the word OLD (forced–choice) or determine whether the pair contained the same (e.g., OLD OLD) or different (e.g., NEW OLD) lexical items. Again, the critical contrast was between the decision types for pairs containing one of each item because in these cases the evidence is matched across the decisions. Despite ceiling levels of accuracy (forced-choice = 98.33, same-different = 95.83, NS), subjects nonetheless slowed considerably during the same–different judgment (1688 vs. 1184 ms; t(11) = 9.97, P < 0.001) and demonstrated increases in PFC regions implicated in the analogous memory contrast, including dorsolateral and frontopolar PFC areas (Fig. 3a and Table 2). Regional overlap was further confirmed using inclusive masking for the memory and lexical decision scan contrasts. As shown in Figure 3b, common dorsolateral and frontopolar regions were implicated across the memory and lexical tasks. Because large differences in reaction time did not modulate these areas during the forced–choice memory contrasts, it is unlikely that these areas identified in the lexical task contrasts merely reflect differences in time on task. Instead, they also appear to reflect a difference in decision operations, which generalizes across memory and lexical task variants.
|
|
To verify that the overlapping dorsolateral and right frontopolar regions were modulated by the decision rule in a similar manner across the memory and lexical tasks, the mean signal from 4 to 6 s poststimulus onset was entered into a rule (forced–choice vs. same–different) by task (memory vs. lexical decision) analysis of variance (ANOVA). For right dorsolateral PFC (Fig. 3b, region 1), this yielded the expected, large main effect of rule (F1,11 = 27.31, P < 0.001), no effect of task (F1,11 = 2.76, P = 0.12), and no evidence for an interaction between the two (F < 1). Similarly, left dorsolateral PFC (Fig. 3b, region 3) yielded the expected main effect of rule (F1,11 = 37.52, P < 0.001), no effect of task (F1,11 = 2.57, P = 0.14), and, importantly, no evidence for an interaction between the two (F > 1). Finally, in the case of right frontopolar cortex (Fig. 3b, region 2), there was a main effect of rule (F1,11 = 13.21, P < 0.005), a main effect of task (F1,11 = 5.06, P < 0.05), and the interaction again was not significant (F1,11 = 4.22, P = 0.064), although it trended in this direction.
Overall, the ANOVAs demonstrate that in all cases there was a prominent and large effect of rule type on the response of the dorsolateral and frontopolar regions that did not require qualification by the presence of significant interaction across rule and task type. Although the main effect of rule type was confirmed, the main effect of task type, which was present in right frontopolar PFC, is difficult to interpret. This is because the mean response for the tasks (memory or lexical) is measured in reference to an implicit baseline, which may vary somewhat across scans. Because the estimated baseline is dependent upon the efficiency of the modeled design, and this design differs somewhat across the scans, differences in the absolute magnitude of responses across different scans are difficult to interpret definitively, and so this effect will not be further considered. This concern does not apply to the main effect of rule type (same–different greater than forced–choice) or the interaction of rule type and task because these are derived from the relative differences in modeled responses within each scan.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and...ResultsDiscussionReferences |
|---|
Overall, these data demonstrate that PFC activations in bilateral dorsolateral and right frontopolar regions during memory retrieval do not directly reflect the level of available memory evidence, the tendency toward successful responding, or the time necessary to execute memory responses. Importantly, all 3 of these behavioral markers have been proposed as key indicators of the levels of retrieval monitoring (Henson and others 2000
; Rugg and others 2003) or executive control required during retrieval attempt (Velanova and others 2003). More specifically, these frameworks assume that reductions in the level of supporting memory evidence (through encoding or retrieval manipulations) will lead to slower, less accurate responding and concomitantly necessitate increased retrieval monitoring of episodic content or increased executive control during attempted retrieval. The key result that weighs against this evidence-based interpretation of PFC retrieval activity was the failure to find activity differences in the comparison of 3AFC versus 2AFC recognition in dorsolateral and right frontopolar regions. Despite the fact there is a clear performance difference across these tasks, dorsolateral and frontopolar responses responded almost identically. In contrast, the comparison of 2ASD versus 2AFC judgments yielded prominent increases in dorsolateral and right frontopolar regions despite the fact that the memory evidence was matched across the judgments. At present, neither the retrieval monitoring nor the executive control framework makes a distinction between evidence- and rule-based modulations of performance, and so the current data pattern is difficult to fully accommodate within these frameworks.
These data are also potentially challenging for another general hypothesis regarding PFC activity during episodic retrieval, namely, the retrieval mode hypothesis (Nyberg and others 1995). Retrieval mode is characterized as a cognitive set or mental state that allows observers to treat probe items as cues to previously experienced target events and enables the subjective experience of remembering should memory content be recovered (Tulving 1983). In contrast to the retrieval monitoring and executive control frameworks, retrieval mode is held to be insensitive to the level of successful retrieval/evidence and should also be insensitive to surface changes in the format of the retrieval tasks provided the mode is maintained for similar lengths of time (Lepage and others 2000; Rugg and Wilding 2000). In a recent multistudy analysis of positron emission tomography research potentially capturing retrieval mode activations, Lepage and others (2000) identified bilateral frontopolar, opercular, and right dorsal prefrontal regions as sites underlying retrieval mode. In the current data, the right dorsolateral and frontopolar regions represent potential retrieval mode candidate sites. However, because retrieval mode is by definition insensitive to the surface characteristics of the retrieval tasks (cf., Fig. 2c) and should not generalize to nonepisodic tasks (cf., Fig. 3a,b), the current dorsolateral and frontopolar regions are clearly not candidates for retrieval mode.
Although these data specifically implicate the dorsolateral and frontopolar areas in domain-general decision operations, their nature is uncertain. However, the use of more formally modeled decision rules, such as those distinguishing forced–choice and same–different judgments, potentially furthers our understanding of the role of these regions. Subsequently, we offer tentative functional characterizations drawing upon SDT considerations. In addition, we contrast these characterizations with 2 hypotheses proposed in the executive control research literature that implicate similar dorsolateral PFC regions.
Borrowing from SDT, the same–different and forced–choice decisions differ in 2 fundamental respects. The first is whether the subject must separately classify the probes or instead makes a relative comparison between them. During same–different judgments, a separate conclusion must be reached regarding each item's category status prior to reaching the final conclusion that the two are either the same or different. During forced–choice, however, such independent classifications are unnecessary, and indeed nonoptimal, because relative evidence is what is diagnostic. That is, it is irrelevant whether both items fall at the low or high end of the evidence scale, merely that one has noticeably more evidence than the other. Thus, under SDT, the relative ease of forced–choice compared with same–different judgments is a by-product of the different decision operations required for each.
The distinction between the relative versus absolute classification requirements of the current tasks appears somewhat similar to executive control research, suggesting a role for dorsolateral PFC in the "free" versus "fixed" selection of stimuli or mental representations. During free selection tasks, subjects attempt to generate a random sequence of selections or classifications from a fixed set of options (e.g., generating a novel sequence of finger taps). During such tasks, functional imaging studies have shown increased activity in dorsolateral regions compared with tasks in which the sequences are fixed (overlearned) or externally directed (Frith and others 1991; Petrides and others 1993). Also, transcranial magnetic stimulation disruption and permanent lesion of dorsolateral regions impair performance more on free compared with fixed selection tasks (Petrides and Milner 1982; Hadland and others 2001). However, in the current study, the "selection" is in fact always free. That is, the decision rule manipulation never directly specifies the status (in terms of memory or lexical identity) or appropriateness of any particular probe. Given this, the free/fixed dichotomy does not seem to adequately capture the key distinction between same–different and forced–choice responding demonstrated here. Another line of research implicating dorsolateral regions during executive control contrasts the manipulation versus simple maintenance of object representations during working memory. The key finding in this literature is that when subjects must perform mental operations upon stimuli, such as sorting them alphabetically, there is an increased activation in dorsolateral regions compared with when the stimuli are passively held in mind during a comparable delay interval (Petrides 1996; D'Esposito and others 1999). Although the lexical and memory classifications performed in the current study seem to clearly require manipulation, it is not clear whether the manipulation hypothesis itself predicts the current outcomes. For example, forced–choice recognition can be thought of as a ranking operation in which the memory evidence associated with each probe is rank ordered prior to selection. Given this, it would be natural to assume that 3AFC demands more manipulation than 2AFC, and the large reaction time difference between these 2 tasks would also suggest this. However, activity in dorsolateral regions was virtually identical across the forced–choice variants. Furthermore, as with the evidence-based hypotheses discussed above, the manipulation hypothesis does not, in and of itself, predict that same–different judgments should yield greater PFC activity than the forced–choice conditions for evidence-matched probes. For such predictions, we contend that more specific decision models such as those within SDT are necessary, and we propose that the dorsolateral activity reflects the SDT prediction that same–different versus forced–choice tasks differ in the number of required classifications that must be executed during the course of the trial.
The second key difference between the forced–choice and same–different tasks revolves around the need, during the latter, to integrate the 2 separate classifications into a final decision. This requirement appears closely linked with several current theories of frontopolar function, particularly those that link frontopolar regions to tasks requiring "subgoal processing" (Fletcher and Henson 2001; Braver and Bongiolatti 2002), internally directed or gated attention (Burgess and others 2005), cognitive integration (Badre and Wagner 2004), or abstract relational reasoning (Christoff and others 2001). Within these frameworks, frontopolar regions are assumed critical under circumstances where current performance depends upon the assessment of multiple intermediate cognitive outcomes or operations. The current data are broadly consistent with these ideas because pairs of episodic or even very simple lexical classifications appear capable of driving frontopolar responses when they must be jointly considered prior to rendering a same or different conclusion.
In summary, these data further our knowledge of the role of dorsolateral and frontopolar regions during memory retrieval by demonstrating that activity in these regions is closely tied to decision operations that generalize across domains (verbal episodic memory vs. lexical decision) and that these operations do not directly reflect the level of evidence supporting the judgments. More generally, the data serve to highlight the distinction between rule-based and evidence-based modulations of discrimination performance and suggest that dorsolateral and frontopolar regions are critically involved in the former and not directly tied to the latter. Consistent with prior research, we suggest that dorsolateral activity is linked to the number of criterial judgments or classifications that must be executed during the course of the trial. Because forced–choice responding utilizes a relative classification rule, only a single item is classified as a target regardless of whether it is a 2- or 3-alternative choice variant. In both cases, the observer is assumed to reserve judgment until an item is identified that satisfies the rule (viz., the item with the most evidence). That is, having determined that a particular item has the most available evidence, and should therefore be classified as old, does not necessitate "backtracking" and additionally classifying the remainder of items as new. In contrast, same–different discrimination requires 2 intermediate item classifications en route to the final decision and therefore results in greater activity in dorsolateral regions. With respect to the frontopolar region, previous research and the current same–different task link frontopolar responses with the need to compare or contrast multiple prior cognitive outcomes prior to rendering a terminal conclusion.
|
References |
|---|
|
TopAbstractIntroductionMaterials and...ResultsDiscussionReferences |
|---|
Badre D and Wagner AD. (2004) Selection, integration, and conflict monitoring; assessing the nature and generality of prefrontal cognitive control mechanisms. Neuron 41:3473–487.[CrossRef][ISI][Medline]
Braver TS and Bongiolatti SR. (2002) The role of frontopolar cortex in subgoal processing during working memory. Neuroimage 15:3523–536.[CrossRef][ISI][Medline]
Burgess PW, Simons JS, Dumontheil I, Gilbert SJ. (2005) The gateway hypothesis of rostral prefrontal cortex (area 10) function. In Duncan J, Phillips L, McLeod P (Eds.). Speed, control and age: in honour of Patrick Rabbitt(Oxford University Press, Oxford) pp. 217–248.
Cabeza R, Dolcos F, Prince SE, Rice HJ, Weissman DH, Nyberg L. (2003) Attention-related activity during episodic memory retrieval: a cross-function fMRI study. Neuropsychologia 41:3390–399.[CrossRef][ISI][Medline]
Cabeza R, Grady CL, Nyberg L, McIntosh AR, Tulving E, Kapur S, Jennings JM, Houle S, Craik FI. (1997) Age-related differences in neural activity during memory encoding and retrieval: a positron emission tomography study. J Neurosci 17:1391–400.
Cabeza R and Nyberg L. (2000) Imaging cognition II: an empirical review of 275 PET and fMRI studies. J Cogn Neurosci 12:11–47.
Christoff K, Prabhakaran V, Dorfman J, Zhao Z, Kroger JK, Holyoak KJ, Gabrieli JD. (2001) Rostrolateral prefrontal cortex involvement in relational integration during reasoning. Neuroimage 14:51136–1149.[CrossRef][ISI][Medline]
D'Esposito M, Postle BR, Ballard D, Lease J. (1999) Maintenance versus manipulation of information held in working memory: an event-related fMRI study. Brain Cogn 41:166–86.[CrossRef][ISI][Medline]
Dobbins IG, Foley H, Schacter DL, Wagner AD. (2002) Executive control during episodic retrieval: multiple prefrontal processes subserve source memory. Neuron 35:5989–996.[CrossRef][ISI][Medline]
Dobbins IG, Khoe W, Yonelinas AP, Kroll NEA. (2000) Predicting individual false alarm rates and signal detection theory: a role for remembering. Mem Cognit 28:81347–1356.[ISI][Medline]
Duncan J. (2001) An adaptive coding model of neural function in prefrontal cortex. Nat Rev Neurosci 2:11820–829.[CrossRef][ISI][Medline]
Duzel E, Cabeza R, Picton TW, Yonelinas AP, Scheich H, Heinze HJ, Tulving E. (1999) Task-related and item-related brain processes of memory retrieval. Proc Natl Acad Sci USA 96:41794–1799.
Fletcher PC and Henson RN. (2001) Frontal lobes and human memory: insights from functional neuroimaging. Brain 124:Pt 5849–881.
Frith CD, Friston K, Liddle PF, Frackowiak RS. (1991) Willed action and the prefrontal cortex in man: a study with PET. Proc Biol Sci 244:1311241–246.
Hadland KA, Rushworth MF, Passingham RE, Jahanshahi M, Rothwell JC. (2001) Interference with performance of a response selection task that has no working memory component: an rTMS comparison of the dorsolateral prefrontal and medial frontal cortex. J Cogn Neurosci 13:81097–1108.
Henson RN, Rugg MD, Shallice T, Dolan RJ. (2000) Confidence in recognition memory for words: dissociating right prefrontal roles in episodic retrieval. J Cogn Neurosci 12:6913–923.
Henson RN, Shallice T, Dolan RJ. (1999) Right prefrontal cortex and episodic memory retrieval: a functional MRI test of the monitoring hypothesis. Brain 122:Pt 71367–1381.
Lepage M, Ghaffar O, Nyberg L, Tulving E. (2000) Prefrontal cortex and episodic memory retrieval mode. Proc Natl Acad Sci USA 97:1506–511.
Macmillan NA and Creelman CD. (1991) Detection theory: a user's guide. (Cambridge University Press, New York).
Nolde SF, Johnson MK, D'Esposito M. (1998) Left prefrontal activation during episodic remembering: an event-related fMRI study. Neuroreport 9:153509–3514.[ISI][Medline]
Nyberg L, Cabeza R, Tulving E. (1996) PET studies of encoding and retrieval: the HERA model. Psychon Bull Rev 3:2135–148.
Nyberg L, Marklund P, Persson J, Cabeza R, Forkstam C, Petersson KM, Ingvar M. (2003) Common prefrontal activations during working memory, episodic memory, and semantic memory. Neuropsychologia 41:3371–377.[CrossRef][ISI][Medline]
Nyberg L, Tulving E, Habib R, Nilsson L-G. (1995) Functional brain maps of retrieval mode and recovery of episodic information. Neuroreport 7:1249–252.[ISI][Medline]
Petrides M. (1996) Specialized systems for the processing of mnemonic information within the primate frontal cortex. Philos Trans R Soc Lond B Biol Sci 351:13461455–1461 discussion 1461–1452.[ISI][Medline]
Petrides M, Alivisatos B, Meyer E, Evans AC. (1993) Functional activation of the human frontal cortex during the performance of verbal working memory tasks. Proc Natl Acad Sci USA 90:3878–882.
Petrides M and Milner B. (1982) Deficits on subject-ordered tasks after frontal- and temporal-lobe lesions in man. Neuropsychologia 20:3249–262.[CrossRef][ISI][Medline]
Ranganath C, Johnson MK, D'Esposito M. (2000) Left anterior prefrontal activation increases with demands to recall specific perceptual information. J Neurosci 20:22RC108.
Rugg MD, Henson RN, Robb WG. (2003) Neural correlates of retrieval processing in the prefrontal cortex during recognition and exclusion tasks. Neuropsychologia 41:140–52.[CrossRef][ISI][Medline]
Rugg MD and Wilding EL. (2000) Retrieval processing and episodic memory. Trends Cogn Sci 4:3108–115.[CrossRef][ISI][Medline]
Slotnick SD, Moo LR, Segal JB, Hart J Jr. (2003) Distinct prefrontal cortex activity associated with item memory and source memory for visual shapes. Brain Res Cogn Brain Res 17:175–82.[CrossRef][Medline]
Slotnick SD and Schacter DL. (2004) A sensory signature that distinguishes true from false memories. Nat Neurosci 7:6664–672.[CrossRef][ISI][Medline]
Tulving E. (1983) Elements of episodic memory. (Clarendon Press, Oxford).
Velanova K, Jacoby LL, Wheeler ME, McAvoy MP, Petersen SE, Buckner RL. (2003) Functional-anatomic correlates of sustained and transient processing components engaged during controlled retrieval. J Neurosci 23:248460–8470.
Wagner TD and Nichols TE. (2003) Optimization of experimental design in fMRI: a general framework using a genetic algorithm. Neuroimage 18:2293–309.[CrossRef][ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  E. Koechlin and A. Hyafil Anterior Prefrontal Function and the Limits of Human Decision-Making Science, October 26, 2007; 318(5850): 594 - 598. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||