R-B Category Learning in PD 1 Characterizing Rule-Based Category Learning Deficits in Patients with Parkinson's Disease
چکیده
Parkinson's disease (PD) patients and normal controls were tested in three category learning experiments to determine if previously observed rule-based category learning impairments in PD patients were due to deficits in selective attention or working memory. In Experiment 1, optimal categorization required participants to base their decision on a single stimulus dimension and ignore irrelevant variation on another dimension, thus emphasizing selective attention processes. In Experiment 2, optimal categorization required participants to base their decision on both stimulus dimensions using a conjunction of unidimensional decisions. Thus, this task placed less emphasis on selective attention and more on working memory. In Experiment 3, optimal categorization again required participants to base their decision on both stimulus dimensions using a disjunction of two unidimensional decisions in which an additional verbal operation was needed, thereby placing even greater emphasis on working memory. Results indicated that PD patients were impaired in the unidimensional rule-based condition, but not the other two rulebased conditions. These results are consistent with previous studies that demonstrate that PD patients are impaired in learning rule-based categories when selective attention demands are greatest, whereas these patients are normal in learning rule-based tasks when working memory demands are emphasized. Overall, these findings help to delineate the conditions under which PD patients display rule-based category learning deficits. R-B Category Learning in PD 3 Characterizing Rule-Based Category Learning Deficits in Patients with Parkinson's Disease A growing body of research indicates that patients with Parkinson's disease (PD) are impaired in their ability to learn new categories (Ashby, Noble, Filoteo, Ell, & Waldron, 2003b; Filoteo, Maddox, Ing, Zizak, & Song, 2005a; Filoteo, Maddox, Salmon, & Song, 2005b; Knowlton, Mangels, & Squire, 1996; Maddox, Aparicio, Marchant, & Ivry, 2005; Maddox & Filoteo, 2001; Shohamy et al., 2004a; Shohamy, Myers, Onlaor, & Gluck, 2004b; Witt, Nuhsman, & Deuschl, 2002a). However, not all studies have identified a generalized category learning deficit in these patients (Filoteo et al., 2005a, 2005b; Maddox & Filoteo, 2001; Peigneux, Meulemans, Van der Linden, Salmon, & Petit, 1999; Reber & Squire, 1999; Smith, Siegert, McDowall, & Abernethy, 1994; Witt, Nuhsman, & Deuschl, 2002b). One potential reason for this discrepancy is the likelihood that there are multiple category learning systems (Ashby & Maddox, 2005; Keri, 2003; Smith, Patalano, & Jonides, 1998), with some of these systems being impacted by the pathological changes to the striatum that occur in PD and other category learning systems not being affected. For example, PD patients' category learning abilities may be impaired to a greater extent when learning is based on trial-by-trial feedback (Filoteo et al., 2005a; Shohamy et al., 2004b), but not when learning is based on simple observation of category exemplars (Reber & Squire, 1999). This difference may be due to PD patients experiencing a deficiency in the dopamine-mediated reward signal that likely drives trialby-trial feedback learning (Aron et al., 2004), but having an intact perceptual priming system that is likely responsible for certain aspects of observational learning (Reber, Stark, & Squire, 1998). Another area that PD patients might experience a deficit is in their ability to learn rulebased categories. Rule-based categories are thought to be learned via an explicit hypothesistesting system that seems to rely on hypothesis generation and testing, logical reasoning, working memory and executive attention. Rule-based categories are those where the rule defining category membership is salient and verbalizable, and can often be based on a single stimulus feature (e.g., the stimulus goes into one category if it is a certain color and another category if it is a different color; Ashby, Alfonso-Reese, Turken, & Waldron, 1998; Bruner, Goodnow, & Austin, 1956; Smith et al., 1998). These tasks are often referred to as unidimensional rule-based tasks because information on a single dimension defines category membership. Perhaps the most wellknown rule-based category learning task is the Wisconsin Card Sorting Test (WCST; Heaton et al., 1981), on which PD patients have been shown to be impaired in a number of previous investigations (Alevriadou, Katsarou, Bostantjopoulou, Kiosseoglou, & Mentenpoulos, 1999; Bowen, Kamienny, Burns & Yahr, 1975; Brown & Marsden, 1988; Caltagirone et al., 1989). Although these studies might suggest a deficit in the learning of rule-based categories, one potential confound with the WCST is that the level of performance is often based on the ability to switch to a new rule after another rule has been established. The index measuring this sort of 'setshifting' on the WCST is the number of perseverative responses (i.e., the number of times a participant responds with a previously correct category although that category is no longer correct), and it is on this index that most studies report impairment in patients with PD. Thus, although PD patients are impaired on this test, their deficit may be due primarily to an inability to switch between rules and not necessarily in rule acquisition (see also Cool, Barker, Sahakian, & Robbins, 2001; Owen et al., 1993b). In a series of recent studies, we examined PD patients' ability to learn rule-based categories. In one study (Maddox & Filoteo, 2001), non-demented PD patients were normal in learning a rule-based task that required participants to compare the length of two lines and categorize the stimuli into one category if the vertical line was longer than the horizontal line, or into another category if the horizontal line was longer than the vertical line. These findings gave the initial impression that PD patients are normal at rule-based category learning. However, in a R-B Category Learning in PD 4 follow-up study (Ashby et al., 2003b) we found contradictory results. In that study, participants were asked to categorize single cards that consisted of colored geometric figures on a colored background. Each stimulus varied from trial-to-trial along four binary-valued dimensions. In the rule-based condition, category membership was defined by the value on a single dimension (e.g., color of the stimuli), thus the task was unidimensional. Interestingly, PD patients were impaired in learning, with fewer PD participants than controls being able to meet a specific learning criterion. Maddox et al. (2005) recently found that PD patients are impaired in learning a unidimensional rule-based task when correct categorization is based on either the distance between two lines or the length of a single line, results that again suggest PD patients are impaired in rule-based category learning. Thus, unlike our original finding in which PD patients were normal in rule-based category learning (Maddox & Filoteo, 2001), two subsequent studies indicate that PD patients are impaired in learning unidimensional rule-based tasks. One possible explanation for these discrepant results is that the various rule-based tasks differ in terms of the presence or absence of irrelevant dimensional variation. That is, in our original study (Maddox & Filoteo, 2001), both of the stimulus dimensions were relevant to category membership (i.e., participants had to base their decision on the length of both lines), so there was no irrelevant dimensional variation. In contrast, in the study by Ashby et al. (2003b), one dimension of the stimulus was relevant, and three dimensions could vary randomly from trial-to-trial, and in the study by Maddox et al. (2005), one stimulus dimension was relevant but the other irrelevant dimension varied from trial-to-trial. Thus, the latter two studies potentially required greater selective attention than did the former study. Indeed, the WCST, on which PD patients are often impaired, also requires that the participant attend selectively to a single stimulus dimension when other irrelevant dimensions vary on a trial-by-trial basis. As such, a selective attention deficit might also underlie their impairment in shifting from one rule to another. To examine the role of selective attention in PD patients' rule-based category learning more directly, we conducted a follow-up study (Filoteo, et al., 2005a) in which we systematically manipulated the selective attention requirements during the learning of a rule-based task. Specifically, participants were administered a rule-based task in which they were presented with stimuli that had 4 binary-valued dimensions (similar to those used in the study by Ashby et al., 2003b) in four different conditions. In each of the conditions, one of the binary-valued dimensions determined category membership, and zero, one, two, or three irrelevant dimensions varied from trial-to-trial. Thus, there was a systematic difference among the four conditions in terms of the degree of irrelevant dimensional variation, and as such, the need for selective attention. PD patients' ability to learn the rule-based categories was impacted to a greater extent than controls as the number of varying irrelevant dimensions increased, suggesting that deficits in selective attention might contribute to the PD patients' rule-based category learning deficit. This characterization of the rule-based category learning deficit in PD is consistent with the observation that these patients are impaired on direct tests of selective attention (Dujardin, Degreef, Rogelet, Defebvre, & Destee, 1999; Filoteo & Maddox, 1999; Maddox, Filoteo, Delis, & Salmon, 1996; McDowell& Harris, 1997; Sharpe, 1990, 1992). In fact, in two previous studies using similar methods as those in the present study, we demonstrated that PD patients were impaired in attending selectively to a single dimension of a two-dimensional stimulus, whereas they were normal in attending to two relevant stimulus dimensions (Filoteo & Maddox, 1999; Maddox et al., 1996). The only difference between the present study and our two previous studies was that, in our previous work, we told the participants the categorization rule prior to the start of the experiment, thereby eliminating the need for learning. Taken together, these findings raise the possibility that a selective attention impairment might contribute to any observed rule-based category learning deficit in patients with PD. These findings might then suggest a specific rule-based category learning deficit in PD, one that emerges when selective attention demands are emphasized. However, it is also known that PD patients are impaired in other cognitive processes known to be important for rule-based category R-B Category Learning in PD 5 learning. As noted above, rule-based category learning tasks not only require selective attention, but can also emphasize working memory processes (Ashby et al., 1998; Smith & Sloman, 1994). Working memory is believed to be important for rule-based category learning because participants must (a) maintain on-line a possible rule that dictates category membership, (b) test this possible rule by categorizing the stimuli, (c) incorporate feedback once a response is made, and (d) maintain on-line rejected rules that did not lead to correct category membership (Ashby et al., 1998; Smith & Sloman, 1994). Deficits in working memory have been reported in PD patients in a number of studies. For example, PD patients are impaired on n-back tasks (Costa et al., 2003), self-ordered pointing tasks (Gabrieli, Singh, Stebbins, & Goetz, 1996), delayed nonmatching to sample tasks (Blanchet et al., 2000), span tasks (Gilbert, Belleville, Bherer, & Chouinard, 2005), delayed eye movement tasks (Armstrong, Chan, Riopelle, & Munoz, 2002; Chan, Armstrong, Pari, Riopelle, & Munoz, 2005), and planning tasks (Owen et al., 1992), to name a few. Thus, it is possible that working memory deficits might also contribute to PD patients' impaired rule-based category learning. However, the extent to which working memory deficits impact PD patients' category learning abilities has not been well-studied. As such, whether PD patients' rule-based category learning deficits are due primarily to selective attention deficits is unknown at this time. The purpose of the present study was to further examine the nature of rule-based category learning in patients with PD and determine whether deficits in selective attention or working memory best account for their previously observed impairments. To this end, we examined the performance of non-demented PD patients in several different experiments in which selective attention and working memory demands on rule-based category learning varied. In each experiment, we administered a rule-based task in which participants were shown single stimuli that consisted of a Gabor patch (see Figure 1) that could vary trial-by-trial in the orientation of the gratings and the spatial frequency of the gratings. Participants were shown a single stimulus and asked to categorize it as a member of Category A or Category B. In Experiment 1, participants were administered a unidimensional rule-based task where correct responding required that they attend to only the spatial frequency of the gratings and base their decision on this dimension while ignoring the trial-by-trial variation on the orientation dimension. Thus, selective attention demands were emphasized in Experiment 1. In Experiments 2 and 3, participants were administered two different rule-based tasks, a conjunctive and disjunctive task, respectively. In both of these tasks, optimal responding required that participants attend to both the spatial frequency and orientation of the stimulus and base their categorization decision on both dimensions. Thus, selective attention demands were minimized. However, the conjunctive and disjunctive rules placed greater demands on working memory. The working memory requirements of the three conditions can be demonstrated by a comparison of the logical expressions needed to describe the optimal rule: Unidimensional Rule-Based Condition: If the spatial frequency of the stimulus is small categorize it as A; if the spatial frequency of the stimulus is large categorize it as B. Conjunctive Rule-Based Condition: If the spatial frequency of the stimulus is large and the orientation is relatively horizontal categorize it as A; if not then categorize it as B. Disjunctive Rule-Based Condition: If the spatial frequency of the stimulus is large and the orientation is relatively horizontal or if the spatial frequency of the stimulus is small and the orientation is relatively vertical categorize it as A; if the spatial frequency of the stimulus is small and the orientation is relatively horizontal or if the spatial frequency of the stimulus is large and the orientation is relatively vertical categorize it as B. Note that the main difference between the unidimensional rule as compared to the conjunctive and disjunctive rules is that the latter two require the participant to maintain in working memory two criteria-one for the orientation dimension and another for the spatial frequency dimension. R-B Category Learning in PD 6 These two criteria have to be maintained across trials while learning occurs, which results in the need for greater working memory capacity as compared to when only a single criterion is used. It is well established that conjunctive and disjunctive rules are considered more complex than unidimensional rules (Salatas & Bourne, 1974; Shepard, Hovland, & Jenkins, 1961) and thus likely require greater working memory or capacity (Kruschke, 1992; Maddox, Filoteo, Hejl, & Ing, 2004b; Waldron & Ashby, 2001). In regard to the difference between the conjunctive and disjunctive rules, the main difference is not the number of criterion needed to be maintained in working memory, but rather the length of the logical expression needed to describe the rule. In particular, the need for the or statement in the disjunctive rule leads to a longer logical expression than the conjunctive rule. Previous research suggests that under many circumstances participants have a more difficult time learning disjunctive rules than conjunctive rules even when the logical rule length is equated, the speculation being that disjunctive rules require more working memory (Bourne, 1974; Haygood & Bourne, 1965; Pazzani, 1991). Importantly, rule-based tasks such as those in the present study are impacted by the performance of a secondary working memory demanding task (Zeithamova & Maddox, in press) and more recent observations from Maddox and colleagues (unpublished data) suggest that, under most circumstances, conjunctive rules are impacted to a lesser degree than disjunctive rules by the performance of a secondary working memory task. These observations suggest that the learning of disjunctive rules likely emphasizes working memory to a greater extent than either conjunctive or unidimensional category structures. Previous studies suggest that unidimensional, conjunctive and disjunctive rule-based tasks are learned using an explicit hypothesis-testing system (Waldron & Ashby, 2001; Zeithamova & Maddox, in press). This assertion is based on past studies that have shown that the learning of these rules are not impacted by manipulations known to affect the learning of implicit categories, such as alterations in the timing of the feedback (Maddox, Ashby, & Bohil, 2003; Maddox & Ing, 2005) or the stimulus-response mappings (Ashby, Ell, & Waldron, 2003; Maddox, Bohil & Ing, 2004a), suggesting that rule-based tasks are not learned using perceptual or procedural-based learning processes. In contrast, rule-based category learning is impacted by manipulations that impact working memory, as noted above. The application of formal models (that will also be applied in the present study) to participants' category learning performance when learning unidimensional, conjunctive and disjunctive rule-based tasks also supports the contention that rule-based category structures are learned via a hypothesis-testing system (Ashby et al., 2003a; Maddox et al., 2003, 2004a; Maddox & Ing, 2005; Zeithamova & Maddox, in press). The key brain regions for the hypothesis-testing system are the prefrontal cortex, anterior cingulate, and the anterior regions of the caudate nucleus; neural structures that underlie hypothesis generation and testing, logical reasoning, working memory and executive attention (Ashby et al., 1998; Ashby & Maddox, 2005; Ashby & O'Brien, 2005; Smith et al., 1998), and previous functional imaging studies have implicated these regions (Filoteo et al., 2005c; Nomura et al., in press; Smith et al., 1998). This frontal-striatal circuitry has been shown to be abnormal in patients with PD and thought to underlie many of their cognitive deficits (Bosboom, Stoffers, & Wolters, 2004; Dubois & Pillon, 1997; Owen, 2004), so it is not surprising that these patients would be impaired in rule-based category learning. Nevertheless, the nature of rule-based category learning is poorly understood in PD. If PD patients' rule-based category learning deficits are specific to tasks in which selective attention demands are emphasized, the patients should be impaired only in Experiment 1. However, if working memory deficits also contribute to PD patients' rule-based category learning performance, then they should also be impaired in Experiments 2 and 3, but more so in Experiment 3 in which working memory demands were even greater.