Delayed Alternation 1 Running head: DELAYED ALTERNATION Prefrontal contributions to delayed spatial and object alternation: A PET study

Delayed alternation tasks are frequently employed as probes of frontal lobe functioning. In order to clarify the neural substrates of delayed alternation performance in humans, we measured regional cerebral blood flow with H20 PET in healthy subjects as they performed delayed spatial and object alternation. Consistent with the monkey lesion literature, increased dorsolateral prefrontal activity emerged during delayed spatial, but not delayed object alternation, whereas orbitofrontal activations emerged in both alternation tasks. The possible cognitive processes contributing to the orbitofrontal and dorsolateral prefrontal involvement in delayed alternation are discussed. Additional activations localized to several non-frontal regions suggesting caution in interpreting alternation deficits in patients as strictly reflecting frontal lobe impairment. Delayed Alternation 3 Prefrontal contributions to delayed spatial and object alternation: A PET study Delayed alternation tasks have been widely employed as probes of frontal lobe functions in both humans and animals. These tasks require subjects to select one of two objects on each trial, with the correct response corresponding to the object or location that the subject did not choose on the previous trial. Such tasks are typically conceptualized as working memory tasks because they require the subject to hold on line and update information on a trial-by-trial basis. Several lesion studies in monkeys implicate the dorsolateral prefrontal cortex (DLPFC) as a critical substrate for performing delayed spatial alternations (DSA) (Mishkin, 1957; Mishkin, Vest, Waxler, & Rosvold, 1969; Butters & Pandya, 1969; Goldman, Rosvold, Vest, & Galkin, 1971; Stamm & Weber-Levine, 1971; Miller & Orbach, 1972). Autoradiographic data in monkeys further confirm the engagement of the DLPFC in delayed spatial alternation (Friedman & Goldman-Rakic, 1994). The deficit in DSA that arises following DLPFC lesions is typically thought to reflect the specific involvement of the DLPFC in spatial working memory (Goldman et al., 1971; GoldmanRakic, 1987). Selective lesions of the DLPFC do not typically impair performance of delayed object alternation (DOA) tasks (Mishkin et al., 1969; Mishkin & Manning, 1978). In contrast, deficits in both DSA and DOA often emerge following lesions to ventrolateral/lateral orbital regions along the inferior convexity (Butters, Butter, Rosen, & Stein, 1973; Miller et al., 1972; Mishkin et al., 1969; Mishkin et al., 1978). Despite their widespread use with animals, and increasing use as neuropsychological probes in humans (Oscar-Berman, Zola-Morgan, Oberg, & Bonner, 1982; Freedman & Oscar-Berman, 1986; Freedman, Black, Ebert, & Binns, 1998; Delayed Alternation 4 Freedman, 1990, 1994; Gross-Isseroff et al., 1996; Zohar, Hermesh, Weizman, Voet, & Gross-Isseroff, 1999), only two published neuroimaging papers have examined the neural correlates of alternation tasks. Gold et al. (Gold, Berman, Randolph, Goldberg T.E., & Weinberger D.R., 1996) reported widespread orbitofrontal (OFC) and dorsolateral frontal activations in a hybrid task that combined elements of alternation and delayed response paradigms. However, the hybrid nature of the task makes it difficult to directly interpret these results with reference to standard DSA and DOA tasks. More recently, we (Curtis, Zald, Lee, & Pardo, 2000) reported a PET study of object and spatial alternation performed with a minimal 1-s intertrial delay. Activations emerged in the medial orbital gyrus, inferior parietal lobule, and right anterior hippocampus during both the object and spatial alternation conditions relative to a sensorimotor control task. This confirmed the involvement of the human OFC in alternation tasks. However, no significant activations emerged in the DLPFC region even during the spatial alternation condition. One possible cause for this lack of activation relates to the brief 1-s delay. Numerous studies in both animals and humans indicate the importance of delay length to the level of prefrontal involvement in various cognitive tasks (Goldman-Rakic, 1987; Fuster, 1989). Almost all studies of DSA in monkeys utilize 5-s delays. Interestingly, Miller and Orbach (Miller et al., 1972) observed that monkeys with DLPFC lesions can perform spatial alternations when there is no delay present, perform poorly when their view of the objects is transiently blocked (for less than a second), and perform even worse when their view of the objects is obstructed for 5-s between each trial. Given these data, which suggest a delay-dependent effect of DLPFC lesions on DSA performance, it seems reasonable to ask whether a DSA task with 5-s delay engages the DLPFC in humans. In order to Delayed Alternation 5 determine the neural correlates of DSA, we asked subjects to perform DSA while rCBF was estimated with H2O PET. For purpose of comparison, a small sample of subjects performed a DOA task while undergoing PET scanning. Methods Human Subjects Ten healthy volunteers ( 7 right-handed males 3 right-handed females, mean age = 24 years, range 20 29) participated in the DSA study. A separate group of 6 healthy subjects (4 right-handed males, 1 left-handed male, and 1 right-handed female, mean age = 31 years, range 21 45) participated in the DOA study. All subjects underwent a computerized psychiatric screening and were excluded if they demonstrated any current or past Axis I psychiatric disorders. Subjects also underwent a brief medical screening interview to rule out a history of neurological disorders. All subjects completed written informed consent approved by the Minneapolis VA Medical Center’s Radioactive Drug Research and Human Subject’s Committees. Alternation Tasks All tasks were presented using an IBM compatible personal computer, with a 37 cm diagonal monitor positioned approximately 50 cm in front of the subject’s eyes. Figure 1 provides a schematic diagram of the DSA and DOA tasks. In both conditions, two 3-dimensional line drawings of objects appeared on a computer monitor. The specific objects were selected because they possessed visual features that were easy to distinguish from each other but did not lend themselves to verbal labeling. To select an object, subjects controlled a cursor with the use of a stylus and touch pad, which functioned like a mouse. The cursor position returned to a central point after each response to prevent the Delayed Alternation 6 use of the cursor’s position from the preceding trial as a cue on the current trial. Automated software recorded all subject responses to determine accuracy and adherence to task demands. Once subjects indicated their response, the objects were left on the screen for an additional 1-s accompanied by the word “correct” (colored green) or “incorrect”(colored red). On the first trial, the subject’s choice was counted “correct” regardless of the subject’s selection. The position of the two objects changed across trials according to a Gellerman randomization schedule (Gellerman, 1993). In both the DSA and DOA tasks, a 5-s delay was interposed between trials during which time the screen remained dark except for a fixation point that subjects were instructed to fixate on at all times. Prior to scanning, all subjects were informed of the rule that governs correct performance, and they practiced performing the tasks to a criterion of five consecutively correct trials. The control condition for both the DSA and DOA consisted of stimuli presented in the exact same format as in the DSA and DOA conditions, except that on each trial a 10-point font asterisk was embedded randomly on one of the two figures. Subjects were instructed to choose whichever object contained the asterisk. Scan order was approximately counterbalanced across the alternation and sensorimotor control conditions. Since one of our motivations was to examine areas becoming active (or showing more activity) with increasing delay demands, we also collected data while subjects performed spatial alternation with a 1-s delay (SA-1s). Nine subjects received both DSA and SA-1s condition (each of these subjects were included as part of our previous report on SA-1s; Curtis et al., 2000). These subjects additionally received a separate sensorimotor control condition with 1-s intertrial delay (as opposed to the 5-s intertrial delay used for the DSA and DOA tasks). Delayed Alternation 7 PET Imaging and Analysis rCBF was estimated from tissue radioactivity using a Siemens ECAT 953B camera (Knoxville, TN) with septa retracted, a slow-bolus injection of H2 O [0.25 mCi/kg infused at a constant rate over 30 s (Silbersweig et al., 1993)], a 90-s scan acquisition, and a 10-minute interscan interval. Scanning began upon radiotracer arrival into the brain, which was timed to occur during the 2 3 trial of the task condition. Images were reconstructed using a 3-D reconstruction algorithm with a Hanning filter (0.5 cycles/pixel)(Kinahan & Rogers, 1989) and were corrected for attenuation with a measured 2-D transmission scan. Measured coincidences were corrected for random detections and electronic dead time, but no corrections were made for decay or scatter. Normalization for global activity (1,000 counts), coregistration within each study session, placement of the intercommissural line from image fiducials, non-linear warping of each subject's scans to a reference stereotactic atlas (Talairach & Tournoux, 1988), and statistical analyses were accomplished with software developed and provided by Minoshima and co-workers (Minoshima, Berger, Lee, & Mintun, 1992; Minoshima et al., 1993; Minoshima, Koeppe, Frey, & Kuhl, 1994). Images were blurred with a 6.0 mm 3-D gaussian filter producing a final image resolution of 10 mm full-width at half-maximum (FWHM). We adopted a significance threshold of p < 0.0001 based on previous studies of the rate of false positive foci arising in bootstrapping analysis (Zald, Lee, Fluegel, & Pardo, 1998). Statistics reflect the maximum magnitude of rCBF increase in a given region relative to the global variance of all pixels within the brain (Worsley, Evans, Marrett, & Neelin, 1993). Identification of cerebellar foci utilized the atlas of Schmahmann et al. (Schmahmann, Doyon, Toga, Petrides, & Evans, 2000), while Delayed Alternation 8 identification of orbital foci followed the nomenclature and probability maps of Chiavaris (Chiavaras, LeGoualher, Evans, & Petrides, 2001) and the cytoarchitectural nomenclature of Ongur and Price (Ongur & Price, 2000). Results All subjects performed the DSA tasks at 100% accuracy. Table 1 displays the results of the contrast between the DSA condition and the sensorimotor control task. Consistent with the lesion literature, performance of the DSA task was associated with significantly increased rCBF in a portion of the dorsolateral prefrontal cortex (middle frontal gyrus, BA 9)(see Figure 2a). As can be seen from Figure 2, this activation represents a discrete focus that appears strictly laterlalized to the left hemisphere. Another discrete focus emerged in the left posterior orbital gyrus (See Figure 2b). Unexpectedly, the largest volume focus localized to the precuneus region (Figure 2c). In two subjects, this focus lies near the edge of the PET camera’s field of view. Because this could result in interpolation errors during 3-D image reconstruction, we reanalyzed the data without these subjects. However, this did not substantially alter the location or magnitude of the focus. Finally, an additional cluster of foci localized to the superior frontal gyrus. The present data suggest performing DSA engages a discrete portion of the DLPFC, whereas previous data did not indicate activation of the DLPFC at a 1-s delay. To more carefully examine this difference, we performed a region of interest analysis comparing activations in the DSA condition (relative to the sensorimotor control condition) with activations emerging during the SA-1s condition (relative to a sensorimotor condition utilizing a 1-s delay). We limited this analysis to 9-subjects who Delayed Alternation 9 completed both the DSA condition and the briefer SA-1s condition, to ensure that differences did not arise as a consequence of using of different samples. A 2-pixel radius spherical region of interest was centered on the peak coordinate of the DLPFC activation in the DSA-sensorimotor control contrast, and the corresponding position in the 1-s spatial alternation-sensorimotor (1-s intertrial delay) contrast. In the DSA condition, rCBF increased by a mean of 8.1% (SD = 5.52 %) across the region of interest, whereas rCBF showed no substantial change (M = -1.21 %, SD = 5.9 %) in the SA-1s contrast. This greater activity occurred despite the fact that there were almost 4 times more trials in the SA-1s condition than the DSA condition, t (8) = 4.39, p < .005. Subjects performed the DOA task with reasonable accuracy (M = 97.0 %, SD = 5.0 %). Table 2 displays the contrast between the DOA and sensorimotor control task. Activations emerged in the left OFC (See Figure 3), anterior insula, supramarginal gyrus and cerebellum. As in our previous study of OA1s, and consistent with data from animal lesions studies, no activations emerged in the DLPFC. Discussion The present data indicate that performance of DSA is associated with activation of the DLPFC in humans. This finding converges with the non-human primate literature, which indicates that lesions of the DLPFC impair the performance of DSA (Mishkin, 1957; Mishkin et al., 1969; Butters et al., 1969; Stamm et al., 1971; Miller et al., 1972), and performance of DSA induces increased glucose utilization in the DLPFC of nonlesioned animals (Friedman & Goldman-Rakic, 1994). Given the greater DLPFC activity in the DSA task relative to the SA-1s condition, it seems tempting to speculate that this activity specifically relates to the greater Delayed Alternation 10 mnemonic demand of the task. On the surface, such an interpretation appears consistent with both the animal literature on DSA (Miller et al., 1972; Goldman et al., 1971) and theoretical conceptualizations about the importance of the DLPFC to spatial working memory (Goldman-Rakic, 1987). Furthermore, Sweeney et al. (Sweeney et al., 1996) observed an activation in a fairly similar region of the left MFG in a spatial oculomotor delayed response task. However, several factors temper this interpretation. First, the left MFG has not emerged as a significant focus of activation in numerous other spatial working memory studies, (see (D'Esposito et al., 1998) for review). This portion of the left middle frontal gyrus or regions proximal to it, have sometimes emerged in tasks involving the monitoring of auditory, verbal, or nonspatial visual information in working memory (Petrides, Alivisatos, Meyer, & Evans, 1993; Zatorre, Evans, & Meyer, 1994; Schumacher et al., 1996; Smith, Jonides, & Koeppe, 1996; Braver et al., 1997). Thus, to the extent that the focus is attributable to working memory, it does not appear to specifically reflect spatial working memory per se. Indeed, it should not be assumed that the activity reflects the maintenance of sensory information (either in terms of a spatial code or a “left/right” verbalization) during the delay period. Instead, the activation may reflect motor preparation, since subjects know the location of the next correct response throughout the entire delay. Thus, to perform the DSA task, one can maintain a retrospective representation of the sensory information (i.e., the object) or a prospective representation of the upcoming motor command. In a study of DLPFC activity during the delay period of delayed response tasks, Quintana and Fuster (Quintana & Fuster, 1992) observed cells in the DLPFC whose activity was linked to the direction of a post-delay motor response [see also (Rainer, Rao, & Miller, 1999)]. In other words, these cells show Delayed Alternation 11 activity related to motor preparation as opposed to holding sensory information on line. The importance of the DLPFC for motor preparation in delayed responses tasks has been confirmed in studies of patients with frontal lesions (Ferreira et al., 1998) and in fMRI investigations with healthy subjects (D'Esposito, Ballard, Zarahn, & Aguirre, 2000; Pochon et al., 2001). The left lateralization of the observed DLPFC activity in the present study might thus relate to a right hand motor preparation in the present study (all subjects responded with their right hand). The motor preparation issue might also explain why a similar DLPFC activation did not emerge in the DOA task, since in the DOA task subjects do not know which direction to move until the objects are represented after the delay period. In considering the present result it may also prove useful to compare the DLPFC result to findings in the Gold et al. (1996) study of a hybrid delayed response alternation task. The observed DLPFC focus in the present study appears quite close to activations reported by Gold and colleagues ( e.g., x = -34, y = 24, z = 28). However, the activation in the present study involved only a discrete area of the frontal lobe, whereas Gold et al. observed widespread activity throughout the frontal lobes in both hemispheres. On the surface this might suggest that Gold’s task produces far more widespread and robust activation of the DLPFC and might make a better probe of frontal lobe functions. However, this speculation is tempered by Gold et al’s observation that activity within the frontal lobe was substantially lower when subjects performed the task a second time. This second performance is more equivalent to the experience level of subjects in our study, since our subjects had already practiced the task prior to scanning. The more restricted nature of delayed alternation activations following practice suggests that patients with Delayed Alternation 12 DLPFC may show deficits in this type of task when the task is novel. In contrast, we would predict that lesions of the DLPFC would not impair performance of alternation tasks once the subject has learned the task rule and practiced applying it. Even lesions encompassing the middle frontal gyrus region, where we observed task related activity in the present study, may not impair DSA because subjects can adopt the alternate strategy of remembering stimulus features (spatial location) instead of utilizing a motor preparation strategy. This reasoning provides a clearly testable hypothesis for studies with patient populations. The lack of DLPFC activation in the DOA condition converges with the animal literature, which indicates that DOA performance remains intact following DLPFC lesions (Mishkin et al., 1969; Mishkin et al., 1978). However, it should be noted that this finding contrasts with an autoradiographic study by Friedman and Goldman-Rakic (1994), in which local cerebral glucose utilization was enhanced in the principal sulcus during both a DSA task and a DOA task. Two notable aspects of the Friedman and Goldman-Rakic study stand out and may help to explain why these authors obtained results that contrast with the lesion literature and the present results. First, the autoradiographic study used a 12-s delay, whereas the lesion literature and the present study used a shorter 5-s delay. Given that we have already provided evidence that delay length (1s vs. 5s) influences rCBF patterns during DOA, we cannot rule out the possibility that delay length plays a role in determining DLPFC recruitment during DOA. However, such a possibility seems inconsistent with much of the current human and nonhuman primate literature on object working memory. The second notable aspect of the Friedman and Goldman-Rakic study relates to the relatively large number of errors Delayed Alternation 13 made by their subjects. Errors on this type of task in monkeys typically involve spatial intrusions (i.e., responses consistent with spatial alternation)(Mishkin and Manning, 1978; Friedman and Goldman-Rakic, 1994). Thus, the high error rates may indicate that the subjects frequently reverted to DSA-like strategies. Such intrusions would clearly contaminate the ability to discriminate the results from those obtained during a DSA task. This stands in contrast to the almost perfect DOA performance of subjects in the present study. The present data confirm the involvement of the OFC in both DOA and DSA, which is consistent with both our previous study using briefer 1-s delays, as well as the monkey lesion literature. A few comments are warranted about the exact location of these activations. First, the foci appear to represent a more medial portion of the OFC than might be expected from the effects of lesions in monkeys which often involve more lateral OFC/inferior convexity regions. Second, the exact location of these foci appears different than in the previously reported 1-s delay study. The reason for this shift remains unclear, but may be the result of individual differences in structural or functional anatomy that are not completely resolved in the process of normalization to a common stereotactic space. Future studies utilizing individual subject analysis of functional MRI data may help to clarify these issues. It is difficult to directly compare the present results to the sparse literature on patients with frontal lobe lesions in humans (Freedman and Oscar-Berman, 1986; Freedman et al., 1998). These studies have examined the error rates of patients while they acquire the task. Because the subjects are not explicitly told the alternation rule in these studies, the deficits may reflect a reasoning deficit, rather than a specific deficit in the Delayed Alternation 14 ability to perform the task. This contrasts with the present PET study in which we examined subjects after they acquired and practiced the task rule. Nevertheless, it is notable that Freedman and Oscar-Berman (1986) observed the greatest deficits in both DSA and DOA acquisition when lesions included the OFC. Freedman et al. (1998) noted DOA acquisition impairments following both ventrolateral/orbitofrontal and more medial frontal (Brodmann areas 10, 24 and 32) lesions. Given the lack of medial frontal activations in the present study, it seems plausible that the more medial frontal regions observed by Freedman et al. (1998) reflect aspects of task acquisition rather than performance of already learned a alternations. The present study does not attempt to decompose the cognitive functions leading to OFC involvement in alternation tasks. Mishkin (1964) speculated that the OFC's contribution to alternation tasks relates to the need to inhibit a prepotent response (which would be to respond to the stimulus that was rewarded on the previous trial). However, once the subject has learned the task (as is the case in the present study), it is not clear that such an inhibition remains necessary. Specifically, the prepotency of the win-stay strategy is reduced after the subject has practiced the alternation strategy. Subjects do not describe having to specifically inhibit an urge to respond to the previously correct response instead of alternating between stimuli. An alternative explanation focuses on the hypothesis that the OFC participates in updating and holding on-line information about the relationship between stimuli and rewards (Zald & Kim, 2001). In many memory tasks, subjects need only hold a representation of whether or not they have seen a stimulus before, or which of two stimuli has ever been associated with a reward. However, in the alternation paradigm, both stimuli are seen an equal number of times and Delayed Alternation 15 rewarded an equal or almost equal number of times. Because of this, the ability to update and hold information on-line regarding the last stimulus-reward pairing may take on particular importance in the performance of alternation tasks. Of course, for this hypothesis to possess explanatory power in the present study requires the assumption that the knowledge of being correct acts as a reward. Such an assumption requires testing, although it is consistent with the recent data implicating the OFC in the coding of abstract reward (O'Doherty, Kringelbach, Rolls, Hornak, & Andrews, 2001). A final hypothesis revolves around the possible role of the OFC in reducing interference. Data from humans with frontal leucotomies indicate that the OFC lesions preferentially affect tests of memory under conditions of high interference (Stuss et al., 1982). Alternation tasks involve high levels of proactive interference, in that irrelevant memories of prior trials other than the immediately preceding trial can compete with the relevant memory of the past trial and interfere with correct performance. Indeed, subjects frequently report that the hardest part of the task is distinguishing their last response from earlier responses. Thus, it may be speculated that the OFC contributes to alternation tasks by reducing the effects of proactive interference. However, until studies examine the role of the OFC in specific task subcomponents such as interference suppression and working memory for reward, it will remain difficult to determine the precise nature of the OFC contribution to alternation tasks. The present findings suggest that marked differences in the neural network subserving performance of alternation tasks occur depending upon the modality and delay period of the task. However, some caution is necessary in drawing this conclusion due the use of different samples of subjects in the DSA and DOA conditions. This Delayed Alternation 16 precluded use of a factorial design that would allow examination of the main effects of delay and modality. However, application of such a factorial design is problematic even if all subjects had completed all of the conditions. Specifically, a confound exists in examining the effect of delay because of the dramatically different number of trials in a scan period. Thus, a failure to observe a focus that emerged in the short delay condition may reflect a decrease signal due to the sparseness of the responses. However, the emergence of new areas such as the DLPFC, which were not seen in the shorter delay trials, likely relate to the increased delay time because they emerge despite the far fewer responses produced during the DSA condition compared to the SA-1s condition. The extent to which non-frontal regions show activations during alternation tasks represents a striking feature of the neuroimaging literature on alternation. Non-frontal activations range from the hippocampus and parietal cortex at short delays, to the precuneus, cerebellum and supramarginal gyrus at longer delays. Although the activation of these areas appears to vary depending upon the specific modality (object vs. spatial) or delay length (1s vs. 5s), their presence suggests a danger in conceptualizing performance on alternation tasks purely as a measure of frontal lobe functioning. Other working memory and “frontal lobe” tasks, although sensitive to frontal lobe lesions, rarely show exlclusive specificity for frontal lobe damage (Anderson, Damasio, Jones, & Tranel, 1991; Grafman, Jonas, & Salazar, 1990; Bondi, Kazniak, Bayles, & Vance, 1993; Reitan & Wolfson, 1994; Reitan & Wolfson, 1995; Dudkin, Chueva, Makarov, & Orlov, 1999). Instead, such tasks typically engage and depend on normal functioning within a distributed network of frontal, posterior, and subcortical brain regions (Chafee & Goldman-Rakic, 1998; Postle, Stern, Rosen, & Corkin, 2000; Friedman & GoldmanDelayed Alternation 17 Rakic, 1994; Cohen et al., 1997; Casey et al., 1998; Chafee & Goldman-Rakic, 2000; Oliveri et al., 2001; Rowe & Passingham, 2001). The present results suggest that the neural basis of delayed alternation involves similarly widely distributed networks. Delayed Alternation 18