Abstract
Abstract
INTRODUCTION
Parkinson’s disease (PD) can lead to significant impairments in executive functions [1]. These have been explained by the fact that dorsolateral prefrontal, orbitofrontal and medial frontal areas are connected to specific regions of the striatum and thalamus via several parallel circuits [2].
Random number generation (RNG) is a paradigm that targets aspects of cognitive control, necessitated by the need to suppress the more habitual and pre-potent response of counting or cycling through the set of numbers, and performance monitoring as well [3]. Typically, RNG requires the generation of “random” series of numbers, e.g. by pressing the number keys of a computer keyboard. The speed of the production is often paced by an external stimulus [4].
Brown et al. [5] reported a rather similar performance of PD and normal controls (NC) but found that NC showed a tendency to count in twos, while PD displayed a bias towards counting in ones. The introduction of a secondary manual tracking task, taking away processing resources from the primary RNG task, reversed the bias seen in NC and exacerbated the bias in PD which suggests that the biased output of the number generation system must be actively suppressed by a limited-capacity process, which is impaired in PD. Robertson et al. [6] tested PD and NC in a verbal RNG task and found that randomness decreased in both groups as generation rate (0.5, 1, 2 and 4 seconds) increased or when participants had to engage in an increasingly demanding secondary task. These effects were more pronounced in the patient group which was argued to be due to an impairment of a limited capacity response selection process in PD, which was thought to be due to an impairment of the Supervisory Attentional System in PD.
To further investigate the notion of an impaired processing capacity in PD, the present study contrasted RNG with ordered number generation (ONG) combined with a secondary so called oddball task. Specifically, we were interested in the event-related brain potential (ERP) correlates of RNG. ERPs are minute task-related voltage fluctuations which can be measured non-invasively with electrodes located on the scalp [7]. Secondary oddball tasks have been used previously to assess the processing capacity left to the secondary task by different levels of difficulty in the primary task [e.g., 8]. In particular, the amplitude of the P3 component has been used to quantify processing capacity [e.g., 9]. The logic is as follows: If two tasks share the same processing resources and the amount of resources taken up by a primary task varies as a function of difficulty of the primary task, the amplitude of the P3 component to target stimuli in the secondary task reflects the processing resources left over for this task. The P3b component in the secondary task can thus be used to monitor the difficulty of theprimary task.
A previous study using the combination of RNG or ONG with an oddball task (involving a particular response to an infrequent target tone occurring in a sequence of standard tones) in a group of young healthy participants [10] revealed that the greater need for attentional resources during RNG was reflected by a smaller amplitude of the P3 component to the target tone of the secondary task but only, when the task was paced at a fast rate (1 tone every 800 ± 100 ms) and not for the slower rate (1 tone every 1300 ± 150 ms). This is in line with behavioral studies using a similar dual-task methodology [11]. Secondly, a left frontal negativity peaking 140 ms after the onset of the pacing stimulus was observed during RNG, whenever the participants produced a random response.
If the limited capacity response selector used to generate the responses in RNG is indeed impaired in PD, we expected to see a more pronounced reduction of the P3 component to targets of the secondary task in PD when combined with RNG compared to ONG. We expected this in spite of the fact that Robertson et al. [6] had used a verbal RNG task, whereas we used a task involving keyboard responses. Verbal tasks might be superior in eliciting random behavior, possibly because in manual responding the spatial arrangement of the keys might lead to habitual response tendencies interfering with habitual responding based on numerical proximity. On the other hand, Baddeley et al. [4] have shown equivalence of manual and verbal response modes.
METHODS
Patients
The study was approved by the local ethics committee. All participants gave their written informed consent.
Twelve right-handed PD patients (7 women, mean age 66.5 years, SD 8.9, mean education 11.4 years, SD 3.1) and twelve matched NC (7 women, mean age 65.7 years, SD 6.5, mean education 12.4 years, SD 2.4) were included. All participants had had no prior exposure to random number generation tasks. Dementia was ruled out by the mini-mental status exam (>28 points). Mean disease duration was 10.4 years (SD 6.8, range 4-25) and mean UPDRS-III was 22.3 (SD 12.9, range 4–49). All patients took L-Dopa (mean 675 mg /d, SD 452), 10 patients also received a dopamine agonist (mean total L-Dopa equivalent dose 858 mg /d, SD 567). Three patients received a COMT-inhibitor (entacapone 600, 1200, 2000 mg /d).
Stimuli and procedure
The participants had a keyboard in front of them which contained the keys 0–9 of the number blockand the space bar. A loudspeaker was positioned 1 meter in front of the participants and was adjusted such that sounds had a volume of 91.5 dB (SPL). During each run 125 frequent standard sounds (635 Hz) and 18 target sounds (435 Hz) were presented inrandom order.
The primary task (number generation) comprised two conditions. Ordered number generation (ONG) required the participants to press the number keys in the canonical order, i.e. 1-2-3-4 etc. They had to press the appropriate key as fast as possible upon the presentation of a standard tone. Whenever an infrequent “target” tone was presented, the participants had to press the “0”-key (secondary oddball task). After such a target stimulus the participants had to start again with ONG commencing with thenumber “1”.
For the random number generation (RNG) condition the participants were instructed to press the keys 1–9 in a random order in response to the frequent standard tones. Research by Baddeley and colleagues [4] has shown that keyboard responses yield similar behavior with regard to randomness as verbal responses. Again, the lower target tones required to press the “0” key. Randomness was explained with the “hat” analogy [12]. Participants were told that ‘As an example of the concept of randomness, suppose we had written the numbers 1 to 9 on pieces of paper and put them into a hat. You take out one piece of paper, call out the number on it and return it to the hat. Then you would reach for another piece of paper and do the same thing. The series of numbers you would call out in that way would be random’ [12, 20].
Both conditions, RNG and ONG, had to be performed with the right and left index finger. Each of 8 runs (4 requiring left responses, 4 requiring right responses, 4 RNG, 4 ONG) comprised 125 standard and 18 target stimuli. The pacer tones occurred every 1800 ms ± 150 ms (rectangular distribution). Between runs, short breaks between 1 and 5 minutes were provided as requested by the participants. For the analysis data from runs requiring left and right-handed responses was combined.
Neuropsychological testing
A brief battery of neuropsychological tests comprising the Wisconsin Card Sorting Test, the subtest digit span from the Wechsler Memory Scale, a Stroop-type color–word interference test and the COWA word fluency test was used. In addition motor performance was tested using a finger tapping test (see Lezak et al. [13] for task descriptions).
Data recording
The electroencephalogram was recorded from 30 scalp channels referenced to the right mastoid (bandpass of 0.01–50 Hz, 250 samples / second). Ocular fixation was verified by recordings of the horizontal EOG. Trials contaminated by eye blinks were detected by vertical electrooculogram. For each participant amplitude criteria for the rejection of blinks were determined individually by measuring the amplitude of 10 eye-blink artefacts in the vertical EOG and Fp1 and Fp2 channels. Thresholds were set to 70 percent of the mean amplitude of the eye-blink artefact. Data was visually inspected for other artifacts (e.g., amplifier blocking). Rejection rates were 12.3 % (SD 8.1) for NC and 13.8 % (SD 7.9) for PD. ERPs were obtained by averaging time-locked to the tone-stimuli or to the motor responses. ERPs were filtered by a 14 Hz digital low pass filter prior to analysis unless mentioned otherwise. Waveforms were quantified by mean amplitude measures which were subjected to analyses of variance. The following time-windows and electrodes were chosen based on previous results [10]: standard stimuli: N1 component, Cz electrode, 120–160 ms; target stimuli: N1 component, Cz electrode, 120–160 ms; N2 component, Cz electrode, 220–320 ms, P3b component, Pz electrode, 380–520 ms. To avoid an undue increase of measures and therefore the problem of multiple testing, ERP analysis was restricted to these4 measures.
Measurement of randomness
According to Ginsburg and Karpiuk [14] there are 3 important factors in the description of randomness: cycling, repetition and seriation. For the present study, we selected 5 different parameters: For cycling the GAP Score [14] was obtained. This score results from calculating the median of the gap between successive occurrence of the 1’s, the 2’s etc. The number of repetitions of the same digits was calculated to yield the parameter REP [14]. To describe seriation, the ascending and descending counting bias in steps of one (e.g., 2-3 or 4-3-2, CS1) and steps of two (e.g., 2–4 or 8-6-4 CS2) as described by Spatt and Goldenberg [15] was assessed. These measures take into account the length of the series. The sequence length was squared to give greater weight to longer sequences.
As the number block of a standard qwertz-keyboard was used for input, it is conceivable that sequences involving neighboring keys on this number block (e.g., 8–5–2) could also reflect habitual responding. We consider this a minor problem as it has been shown that numbers are arranged mentally along a mental number line in a quasi spatial fashion [16] from left (small numbers) to right (higher numbers). Moreover, our participants were mostly retired and had little exposureto computers.
RESULTS
Neuropsychological findings
Neuropsychological test results are shown in Table 1 together with the appropriate statistics. The comparison of PD and NC revealed differences in motor speed for both hands. Differences were also found in the COWA, the Stroop-test, and the maximum digitspan forward.
Behavioral results
An increase in the CS1 measure indicated the expected decrease in randomness for the PD patients (Table 1). Moreover, the difference in the REP score approached significance with PD patients showing a more pronounced tendency towards repetition avoidance. There were no group differences with respect to the CS2 and GAP measures.
Table 2 summarizes behavioral data. Response times were shorter for NC than for PD patients (group: F(1,11) = 81.3, p < 0.001), for the ONG compared to the RNG task (task: F(1,11) = 27.3, p < 0.001) and for standard compared to target sounds (stimulus: F(1,11) = 97.2, p < 0.001). There was a group x task (F(1,11) = 8.9, p < 0.02) as well as a group x stimulus x task interaction (F(1,11) = 6.8, p < 0.05). When the standard sounds were examined separately with a group by task ANOVA, an interaction was obtained which was driven by the fact that there was a disproportionate lengthening of response times in the RNG tasks for the PD patients (F(1,11) = 10.7, p < 0.01). With regard to correct button presses there were no significant differences between groups.
Stimulus-locked ERP responses to standard stimuli
The stimulus locked ERPs to the standard stimuli (Fig. 1) show a large negative N1 component with a peak latency of about 140 ms. This was larger for NC but did not show a difference as a function of task (mean amplitude 120–160 ms, electrode Cz, main effect group F(1,11) = 5.5, p = 0.04, main effect task: F(1,11) = 0.95, n.s.). This was followed by a relative negativity peaking around 300 ms which was more prominent for PD (mean amplitude 250–400 ms, group F(1,11) = 6.9, p < 0.025) and the RNG task (F(1,11) = 7.4, p < 0.02). There was no interaction between these two factors.
For RNG standard tones followed by a repetition response (i.e. the subject pressed the same key as in the immediately preceding trial), and standard tones followed by a “random” response (e.g. “8” after “5”) were computed (Figure S1, supplementary materials). ERPs to the “random” trials showed a significantly bigger amplitude than those to repetition trials (F(1,11) = 7.8, p < 0.02). A group x random/repetition interaction (F(1,11) = 4.9, p < 0.05) reflected the smaller difference between repetition and random trials inPD patients.
ERPs to target stimuli
The ERPs to target stimuli are characterized by a large negativity (N1) at about 140 ms which had a higher amplitude in the NC (Fig. 2). This is followed by a second negativity (N2), which is considerably less pronounced in the PD group. Within the PD group the N2 appeared to be smaller in the ONG condition. A positive peak with a centroparietal maximum and a peak latency of about 400 ms (P3b) followed the N2. In the PD group, the P3b was sensitive to task being much smaller in the RNG condition.
As for the standard stimuli, analysis of the N1 component showed a main effect of group (mean amplitude 120–160 ms, electrode Cz, F(1,11) = 6.9, p < 0.05) but neither an effect of task (F(1,11) = 0.51, n.s.) nor an interaction (F(1,11) = 1.52, n.s.).
For the N2 component (mean amplitude 220–320 ms, electrode Cz) the ANOVA revealed a highly significant main effect of group (F(1,11) = 26.3, p < 0.001), but no task effect, whereas the group x task interaction showed a trend (F(1,11) = 4.4, p = 0.059).
The P3b component showed no effect of group (F(1,11) = 3.9, n.s.) but a main effect task (F(1,11) = 7.2, p < 0.05), and a group x task interaction (F(1,11) = 11.3, p < 0.01).
DISCUSSION
The present study investigated electrophysiological correlates of random number generation as a marker of executive functioning in PD. Previous studies of RNG in PD have revealed mixed results: Eberspach et al. [17] in an auditorily paced number generation task found that PD patients showed intrusion of unwanted systematic strategies such as counting suggesting a decreased ability to generate random movement sequences. Using a verbal RNG,Robertson et al. [6] found patients to show a more pronounced decrease of randomness of responses compared to controls as generation rate increased. Brown et al. [5] found relatively similar behaviour in PD and NC in a paced random number generation task (1 response per second) with patients showing a bias towards counting in ones whereas NC showed a bias of counting in twos. Adding a manual tracking task as a secondary task led to a more pronounced bias of counting in ones in the patients. Recently, Anzak et al. [18] have revealed a first clue regarding the pathophysiology of RNG in PD: They recorded local field potentials (LFPs) from depth electrodes located in the subthalamic nucleus (STN) while PD patients either performed paced RNG or ONG tasks. RNG led to a significant increase in the 45–60 Hz gamma band relative to ONG which was correlated to the number of repetitions. Anzak et al. [18] therefore suggested that gamma activity in the STN is relevant for controlledprocessing, in particular the active selection of the same number on successive trials. It is conceivable that STN gamma activity is related to activity changes in cortical areas connected to the STN.
The behavioral results of the current experiments were similar to Brown et al. [5] as the PD patients showed a significant bias towards counting in ones. Also, we found a tendency towards more pronounced repetition avoidance in the PD patients. For the purpose of ERP recording the experimental conditions had to be modified. In contrast to other paradigms the participants did not have to produce numbers while synchronizing to a fixed pacing signal but rather after an acoustic signal was given. This decoupling of pacer tones and responses was achieved by the instructions, a jitter in the interval between two successive tones and finally by the secondary oddball task, which precluded preparation of a number response as the target tone would require a“zero” response.
The current experiment had a dual task character as the tone sequence used as pacer stimulus also contained occasional oddball tones that had to be responded to by pressing the “0” key. With regard to the behavioral indices, the RNG task proved to be more demanding in the PD patients, as response times to the standard tones were disproportionately lengthened in PD patients (RNG/ONG response time difference 100 ms) compared to NC (67 ms).
The dual task methodology is often employed to test demand for attentional resources by different levels of the primary task. In ERP research the amplitude of the P3b component to the secondary task (here: the oddball task) has been used as an index for attentional resources “left over” by the primary task [8]. While P3b amplitude has been known to vary as a function of a number of factors [19], the relation to attentional resources, or in other words processing capacity, appears particularly strong [for a review: 9]. In the present study NC showed a similar amplitude for the target P3b in the ONG and RNG blocks. By contrast, the P3b was considerably smaller in the RNG compared to the ONG blocks in the PD patients. In a previous study in young volunteers, we had found a reduced P3b amplitude for the RNG as well, but only in the fast-paced (one tone every 800 ms) and not in the slower-paced (one tone every 1300 ms) version. This suggested that attentional resources were sufficient to carry out both tasks without any tradeoff at the slower pace, but that at the faster pace significant resources were drawn away from the oddball task in the RNG condition indicating aresource limitation.
Interestingly, even though the present version of the task used a rather slow pacing rhythm (one tone 1800 ms), a P3b difference between RNG and ONG blocks was seen in the PD patients but not in NC. Applying the same reasoning as in Joppich et al. [10] this indicates that resource limitations are already present in PD at this comparatively low pace.
Following Baddeley and colleagues [4] the reduced P3b in the secondary task likely reflects the greater need of guided activation in the RNG condition, thus leaving only limited resources for the secondary task. Alternatively, Brown et al. [5] suggested that during RNG subjects have to control for and suppress highly associated responses (such as counting upwards, 1–2–3 …). Whatever the exact nature of the resource-demanding process in RNG is, it is clear that dual task methodology employed in the present study is sensitive to resource limitations in PD.
In a PET study Jahanshahi et al. [20] found marked differences of task-related activation of DLPFC between conditions with pacing stimuli at intervals of 0.5 seconds or 1.5 seconds. They concluded that an interval of 0.5 s between responses might be too fast to keep up random behavior and that the DLPFC cannot control or select correctly the responses under these conditions, thus favoring habitual, counting behavior. This suggests that activation-level of the DLPFC, as the P3b component in the current study might be used as an index of processing resources taken up by the RNG task.
Limitations
The limitations of the study need to be acknowledged. We conducted this study on a convenience sample of PD patients that was under different levels of dopaminergic medication. Thus, it is impossible to distinguish effects of the disease itself from effects of the medication on RNG. Previous studies on the effect of dopaminergic medication on ERPs in PD has led to inconclusive results. Most data concern the P3 component which has also been examined in the present study. Some studies have found smaller P3 amplitude off medication compared to the on state [21]. Also, some studies have found reduced P3 amplitude in the off state compared to healthy controls [22]. Interestingly, other studies have reported larger P3 amplitude during medication off compared to either medication on [23] or to a group of age-matched control participants [24]. Finally, some studies found no differences as a function of dopaminergic medication [25]. We therefore believe that dopaminergic medication cannotexplain the present pattern of results but a study of RNG on unmedicated de novo patients would be helpful to address this issue. The sample size of 12 in the current study is rather modest and can thus be seen as a further limitation. Finally, it has been shown recently that practice might influence random number generation [26]. In particular, it has been shown training in the RNG task improves the ability to suppress habitual responses as reflected by measures of seriation. It might thus be asked whether differential training effects might explain the differences between groups. This, however, is highly unlikely, as we used RNG and ONG tasks intermixed thus counteracting learning over a session.
CONCLUSIONS
The present data set attests to the utility of electrophysiological measures to reveal the engagement of different portions of the frontal cortex in the generation of random sequences. The P3 modulation in the target ERPs suggests that RNG in comparison to ONG draws more heavily on attentional resources. Further studies are needed to confirm and extend these findings.
CONFLICT OF INTEREST
The authors declare that there is no conflict ofinterest, financially or otherwise.
Footnotes
ACKNOWLEDGMENTS
The work was supported by grants from the DFG and the BMBF to TFM. RD is supported by a grant of the Peter-Max-Müller-Foundation.