The functional locus of the lateralized readiness potential

Transcription

1 Psychophysiology, 41 (2004), Blackwell Publishing Inc. Printed in the USA. Copyright r 2004 Society for Psychophysiological Research DOI: /j x The functional locus of the lateralized readiness potential HIROAKI MASAKI, a NELE WILD-WALL, b JÖRG SANGALS, b and WERNER SOMMER b a Japan Society for the Promotion of Science, Tokyo, Japan b Humboldt-University, Institute for Psychology, Berlin, Germany Abstract The lateralized readiness potential (LRP) is considered to reflect motor activation and has been used extensively as a tool in elucidating cognitive processes. In the present study, we attempted to more precisely determine the origins of the LRP within the cognitive system. The response selection and motor programming stages were selectively manipulated by varying symbolic stimulus response compatibility and the time to peak force of an isometric finger extension response. Stimulus response compatibility and time to peak force affected response latency, as measured in the electromyogram, in a strictly additive fashion. The effects of the experimental manipulations on stimulus- and response-synchronized LRPs indicate that the LRP starts after the completion of response-hand selection and at the beginning of motor programming. These results allow a more rigorous interpretation of LRP findings in basic and applied research. Descriptors: Lateralized readiness potential, Response selection, Motor programming, Functional locus The lateralized readiness potential (LRP) is regarded as a useful and powerful tool in the study of human information processing. It is extracted from event-related brain potentials (ERP) and considered to reflect the activation of response-related processes following stimulus-related processing (Coles, 1989; De Jong, Wierda, Mulder, & Mulder, 1988; Gratton, Coles, Sirevaag, Eriksen, & Donchin, 1988). In previous research, the LRP has been successfully employed to investigate, for example, information transmission between perception- and response-related processes (e.g., Coles, Gratton, & Donchin, 1988; Miller & Hackley, 1992), motor programming (e.g., Leuthold, Sommer, & Ulrich, 1996), overlapping task processing (e.g., Osman & Moore, 1993), or dual route processes in spatial stimulus response compatibility (e.g., Stu rmer, Leuthold, Soetens, Schröter, & Sommer, 2002). Although the somewhat global notion of LRP onset as marker of response activation is quite sufficient for many issues in cognitive psychology, there are other applications where a higher degree of precision as regards the functional significance of this component would be most welcome. For example, Sommer, Leuthold, and Schubert s (2001) assessment Portions of this article were presented at the 42nd Annual Meeting of the Society for Psychophysiological Research, Washington, DC, October 2 6, This study was supported by a fellowship of the Japan Society for the Promotion of Science to the first author. The authors would like to thank Allen Osman for valuable discussions, Kathrin Pusch for her assistance in collecting the data, and William Gehring, Toby Mordkoff, Birgit Stürmer, and one anonymous reviewer for their helpful comments on an earlier draft of this article. Address reprint requests to: Dr. Hiroaki Masaki, School of Sport Sciences, Waseda University, , Mikajima, Tokorozawa, Saitama, Japan, masaki@waseda.jp. of a motoric bottleneck after response initiation (De Jong, 1993) depended crucially on the locus of the LRP after or during response selection but prior to response initiation. The present study attempts to provide further information on how the onset of the LRP is related to the cognitive and motoric processes that have been suggested during choice reaction time tasks. The LRP is derived by recording ERPs from above the motor cortices in tasks that call for left- and right-hand or -foot responses. The ERP above the cortex ipsilateral to the effector required in a given trial is subtracted from the contralateral ERP. When these difference waves are averaged across hands, they yield the LRP, reflecting pure hand-related ERP asymmetry. In general terms, the LRP is considered as a measure of response activation or preparation (cf. Coles, 1989). The interval between a stimulus and the onset of the stimulus-synchronized LRP (stimulus-lrp interval) is a relative measure for the duration of premotoric processes, including perception and at least some aspects of response selection. In contrast, the interval between the onset of the response-synchronized LRP and the response (LRP-response interval) is a relative measure for the duration of the subsequent, that is, the motoric processes (Osman, Moore, & Ulrich, 1995). The LRP is considered to be generated at least partially in the primary motor cortex (e.g., Coles, 1989; Miller & Hackley, 1992). Note however, that even if LRP were exclusively generated in M1, it would, according to current neuroanatomical knowledge, not unambiguously specify its meaning, because the precise functions of M1 are themselves a matter of debate (e.g., Graziano, Taylor, & Moore, 2002). To determine the functional significance of ERP components, it has been suggested to assess the effects of experimental factors with known locus of action within the information processing system on the ERP component in question (e.g., Meyer, Osman, 220

2 Functional locus of the LRP 221 Irwin, & Yantis, 1988). If, for example, manipulation of an experimental factor increases the interval between the stimulus and the onset of the component in question but not the interval between the onset of the component and the response, it can be concluded that the component is elicited in a stage or in stages following the locus of action of the experimental factor. For choice reaction time tasks, Sanders (1980, 1990, 1998) proposed three perceptual stages (preprocessing, feature extraction, and identification) and at least three response-related stages (response selection, motor programming, and motor adjustment). Of primary interest in the present context are responserelated stages. The first one is the so-called response selection or decision stage. In this stage, the stimulus code extracted during the perceptual analysis is assigned to an abstract response code. The manipulation most unambiguously affecting the duration of stimulus response assignment is the compatibility between stimuli and responses (S-R compatibility). Most clearly this holds true for so-called symbolic S-R compatibility, for example when letters or words, meaning left or right have to be assigned to responses performed at the side designated by the stimulus (compatible assignment) or at the opposite side (incompatible assignment). Less well suited for present purposes are variants of spatial S-R compatibility tasks, including the Simon task (Simon & Rudell, 1967) because here, a second, unconditional or direct, processing route may come into play (Kornblum, Hasbroucq, & Osman, 1990), undermining the research logic suggested by Meyer et al. (1988). Following the response selection stage, a motor programming stage has been suggested. Here, a more detailed specification of the kinetic parameters of the response takes place that was established during response selection (Zelaznik & Franz, 1990). Proposed experimental variables related to the motor programming stage are, for example, crossed hand placement, movement direction, and movement velocity (Sanders, 1998). The third motor-related process is the so-called motor adjustment stage that deals with the transition from central to peripheral motor activity. It is considered to be affected by foreperiod duration and instructed muscle tension. In screening the available reports about experimental effects on the onset of the LRP for the sake of interpreting its functional significance, only a limited set of designs is useful. In general, almost all experiments have to be ruled out that have used the LRP as a tool to identify the effects of experimental factors on information processing. In these studies, at least a rough idea about the functional significance of the LRP is a prerequisite and gaining knowledge about the locus of the experimental effects is the aim. In contrast, the present study is concerned with specifying more closely the significance of the LRP by using experimental effects with a well-known locus of action. In addition, we also have to exclude all designs that might induce parallel or dual route processing because they do not allow an unambiguous interpretation of experimental effects on LRP onset. In particular, we cannot take into consideration studies that used multiple stimuli or stimuli with several attributes. This concerns, for example, most studies investigating information transmission (e.g., Gratton et al., 1988; Miller & Hackley, 1992; Osman, Bashore, Coles, Donchin, & Meyer 1992; Smid, Mulder, & Mulder, 1987), spatial S-R compatibility (Wascher, Reinhard, Wauschkuhn, & Verleger, 1999), and the Simon effect (e.g., Stürmer et al., 2002). The studies that can be taken into consideration because they manipulated experimental factors with a commonly accepted locus of action in unambiguous single-route processing are astonishingly few in number. In a two-choice task using the words left or right, Smulders, Kok, Kenemans, and Bashore (1995) factorially manipulated stimulus quality (clear vs. degraded) and response complexity (single keystroke vs. three keystrokes with index, ring, and middle fingers); stimulus quality was taken to affect the feature extraction stage and response complexity was considered to relate to motor programming. As expected, the two experimental factors had additive effects on reaction time (RT). Stimulus degradation increased the stimulus-lrp interval but not the LRP-response interval; conversely, response complexity increased only the LRP-response interval, but not the stimulus- LRP interval. The authors concluded that the LRP succeeds the start of response choice but precedes the start of response programming. On the other hand, the authors noted that the LRP might also emerge within rather than after the stage of response choice. Another problem with this study is that the association of the factor response complexity with the response or motor programming stage is not generally accepted. As several studies suggested (e.g., Verwey, 1994), a sequence of key presses may cause the insertion of additional motoric processing stages into the system rather than merely increasing the time demands for the motor programming stage. This assertion also makes the findings about larger LRP amplitudes for complex as compared to simple responses by Hackley and Miller (1995) and Stief, Leuthold, Miller, Sommer, and Ulrich (1998) difficult to interpret for present purposes. Smulders (1993, chapter 7) manipulated stimulus quality in the same way as Smulders et al. (1995) in combination with symbolic S-R compatibility. Both factors affected the stimulus- LRP interval; the results about the LRP-response interval were not quite as clear because this interval was increased by degraded incompatible stimuli. Miller and Ulrich (1998) reported a manipulation of symbolic S-R compatibility. Participants had to make six-choice responses with index, middle, and ring fingers of the right and left hands. Whereas the choice of hand was always symbolically compatible and easy, the assignment of the stimuli to the fingers within each hand was either easy (compatible) or difficult (incompatible); this within-hand compatibility manipulation affected the LRPresponse interval but not the stimulus-lrp interval, indicating a locus for the LRP onset after the termination of hand selection but prior to finger selection. A similar observation was made by Hackley and Valle-Inclán (1999), who manipulated the number of fingers to be selected within a hand; the number of alternatives affected both the stimulus-lrp and LRP-response intervals. Whereas the former result is compatible with the notion that number of alternatives has effects on both perception and response selection (Sanders, 1998), the latter finding conforms with the one by Miller and Ulrich (1998) that finger selection within a hand follows LRP onset. An additional useful observation is the increase of LRP amplitude and the reduction of the LRP-response interval with the number of prespecified movement parameters in movement precuing tasks (Leuthold et al., 1996; Sangals, Sommer, & Leuthold, 2002; Ulrich, Leuthold, & Sommer, 1998). These findings argue for effects of motor programming processes on the LRP but do not specify the initiation point of the LRP. Together, because of the effects of stimulus degradation and S-R compatibility on the stimulus-lrp interval, it appears rather safe to conclude that the LRP is generated postperceptually and also after at least a considerable amount of response selection. In addition, based on the precuing effects on LRP

3 222 H. Masaki et al. amplitude, it seems that the LRP emerges prior to or during motor programming. In other words, the earliest plausible locus of the LRP onset is during response selection and the latest locus is during motor programming; LRP might also emerge after completion of response selection and at the beginning of motor programming. Because of the residual ambiguities in the results of previous studies, it was the aim of the present experiment to provide more specific information about the relationship of LRP onset to the response selection and the motor programming stages. To determine the locus of the LRP onset, the present study selectively manipulated in a factorial design the time demands for the response hand selection and the motor programming stages. Additivity of the reaction time effects of these manipulations would demonstrate their selectivity. To manipulate response hand selection, we chose symbolic S-R compatibility, as used by Smulders (1993). The choice of the factor acting on motor programming was more difficult. Because of the ambiguities in the interpretation of response complexity mentioned above, we adopted instructed movement velocity in order to manipulate motor programming. Evidence that this variable selectively affects motor programming comes, for example, from studies of the RTs for sliding arm movements by Spijkers and coworkers (Spijkers, 1987, 1989; Spijkers & Sanders, 1984; Spijkers & Steyvers, 1984). Instructed speed of the sliding movement reliably affected RT (shorter RT for faster movements); these effects were additive to those of S-R compatibility and foreperiod duration, segregating the effects of speed manipulation from those presumably affecting response selection and motor adjustment, respectively. In the present study, the S-R compatibility manipulation was realized by presenting the letters R and L, requiring isometric responses (force pulses) with the left and right middle fingers that were either compatible (R! right; L! left) or incompatible (R! left; L! right). Instructed speed was manipulated by requiring in a separate condition that the time to peak force should be within an early, narrow time window or within a later, wider time window; peak force was to be kept within specified and constant limits. Although movement velocity is operationalized here differently than done by Spijkers and coworkers, we take it to be a valid manipulation of the motor programming stage. According to the parallel force unit model proposed by Ulrich and Wing (1991), brief force pulses at the lower limit of time to peak force (i.e., about 100 ms) have only one degree of freedom, that is, only the number of force units has to be defined by the motor program. On the other hand, slower force pulses require the programming of both the number of force units and their timing to obtain the correct force curve. Therefore, the short time-to-peak-force conditions should require less motor programming than long time-to-peak-force conditions. Given that S-R compatibility and movement velocity have independent effects on reaction time, we can make diagnostic predictions for the LRP depending on the functional locus of the origin of this component. The predictions primarily concern the effects of the S-R compatibility and the time-to-peak-force manipulation on the stimulus-lrp interval and the LRPresponse interval. As far as the S-R compatibility effects are concerned, these onset-related predictions can be supplemented by predictions about the slopes of the LRP. However, because it cannot be excluded that time to peak force may affect the amplitude of the LRP (see below), slope-related predictions were prudently confined to effects of S-R compatibility. In principle, five possible loci can be distinguished with the present design. Two of these appear to be highly unlikely on the basis of previous findings. Thus, we do not expect the LRP to start at the beginning of response hand selection, in which case there would be effects of both S-R compatibility and instructed velocity on the LRP-response interval but none on the stimulus- LRP interval. Likewise, we do not expect the LRP to start only after motor programming, which would be indicated by exclusive effects of both variables on the stimulus-lrp interval but none on the LRP-response interval. This leaves us with three realistic alternatives. First, if the LRP starts immediately after the response selection stage and at the beginning of motor programming, the S-R compatibility manipulation should only affect the stimulus-lrp interval with longer onset latencies in the incompatible conditions; also, the slopes of the stimulus- and response-synchronized LRPs should be unaffected by S-R compatibility because the activation processes should be postponed, but their time course should remain the same. Conversely, the time-to-peak-force manipulation should only affect the LRPresponse interval, showing longer intervals in the slow time-topeak-force conditions. Second, if the LRP starts somewhere during response selection, one should expect any manipulation that increases the total time demand in this stage to also increase the time preceding the elicitation of this component; it should, however, also increase the interval between the onset of the component and the execution of the response. Therefore we should expect both the stimulus-lrp and the LRP-response intervals to be longer in the incompatible than in the compatible conditions. The time-topeak-force manipulation, however, should only affect the LRPresponse interval but not the stimulus-lrp interval. Just as one should predict an effect of S-R compatibility on the LRPresponse interval, one might also expect more gradual slopes of both stimulus- and response-synchronized LRPs in the incompatible condition. A similar kind of reasoning holds true for the third alternative, namely that the LRP emerges not at the beginning but during motor programming. In this case, S-R compatibility effects should be confined to the stimulus-lrp interval. Time to peak force, however, should affect both the stimulus-lrp interval and the LRP-response interval. The time-to-peak-force manipulation employed in the present study also offers an opportunity to consider an issue that has yielded controversial results. This issue concerns the influence of kinetic response parameters on LRP amplitude. Sommer, Leuthold, and Ulrich (1994) had required unspeeded responses with short or long time to peak force (100 vs. 200 ms on average) and peak forces of 10 or 50% of maximal force. No effects of the kinetic factors on LRP amplitude were found. Recently Mordkoff and Grosjean (2001) replicated these results in a speeded task by sorting responses according to kinetic parameters. In contrast, the force growth rate had a strong effect on LRP amplitude in an unspeeded task where isometric responses of long duration were compared (Ray, Slobounov, Mordkoff, Johnston, & Simon, 2000). One possibility to explain these results is the difference in time-to-peak-force values between the studies of Sommer et al. and Mordkoff and Grosjean on the one hand and Ray et al. on the other hand. Possibly, time to peak force affects LRP amplitude only if the movement cannot be completely programmed in advance but has to be regulated during its execution. The present study uses speeded responses

4 Functional locus of the LRP 223 with either short-duration ballistic movements and longer duration ramp movements that might require online regulation. If the hypothesis above is correct that LRP amplitude is modulated by online-regulation of the ongoing movement, we should expect LRP amplitude to be larger in the long time-topeak-force condition. Finally, we would like to make a comment on method. One intrinsic problem of the present study is its reliance on onsetmeasures on the one hand and the use of a movement condition with a slowly rising force ramp on the other hand. Clearly, onsets can be determined most easily and most precisely when the rising flank of the parameter is steep. To safeguard against problems in determining force onsets, we required finger extensions and recorded the electromyogram from the musculus extensor digitorum to use it as an alternative indicator of reaction time and also for calculating response-synchronized LRPs (e.g., Masaki, Takasawa, & Yamazaki, 2000). Using the EMG for calculating response-synchronized LRPs has the additional advantage of eliminating any variability that occurs after the start of initial peripheral movement execution. Method Participants Twelve participants (4 men, 8 women), between 19 and 28 years of age (mean SD years), were recruited from Humboldt-University population. All participants were righthanded with mean handedness scores of 181 (Oldfield, 1971) and had normal or corrected-to-normal vision. Informed consent was obtained. Stimuli and Response Measurement The letters R or L subtending approximately served as imperative stimuli. They were presented in white at the center of a computer monitor placed 100 cm in front of the participant. Responses were measured by means of force-sensitive keys. Leaf springs ( mm) were held by clamps at one end, whereas the other end remained free. The thimblelike finger holders were attached on top of the free end of each spring. Strain gauges (Type 6/120 LY 41, Hottinger Baldwin Messtechnik) were attached near the fixed end. Any force applied to the leaf spring via the finger holders at the free end was reflected by an analogous signal, which was digitized along with the electrophysiological signals (see below). The keys allowed for near isometric recording of the required middle finger extensions. Procedure Participants comfortably rested both forearms and palms on flat boards and placed their middle fingers into the holders of the force-sensitive keys. Each trial began with the presentation of a fixation cross for 0.5 s, after which it was replaced by one of the letters, remaining in view until a response occurred. In the compatible conditions, participants responded to the letters L and R by lifting the left or right middle finger, respectively. In the incompatible conditions, the assignment of letters and fingers was reversed. Whereas peak force could always vary between 1.5 and 7.5 N, time to peak force was limited to a range between 70 to 140 ms in fast and 210 to 420 ms in slow conditions. One second after the response, visual feedback was given if time to peak force had been too short ( Zu rasch ) or too long ( Zu langsam ), if peak force had been too weak ( Zu schwach ) or too strong ( Zu stark ), or if the incorrect finger had been used ( Falsche Hand ). Intervals between a response and the next fixation point ranged between 2 and 3 s. Prior to the experiment proper, each participant practiced the different responses for each hand in compatible conditions. In each condition, consecutive blocks of 30 trials each were conducted until performance reached 75% correct in that block. In addition to the various types of error feedback described above, confirmatory feedback was given in the training session whenever the response was correct, consisting of the word OK. Only those participants who reached criterion in all conditions of the training session were used in the experimental session to follow on the next day with one exception where one day elapsed between sessions. The experimental session consisted of eight blocks of 60 trials each with the first 6 trials serving as warm-up after the short rest separating the blocks. The four conditions, consisting of the factor combinations of compatibility (compatible, incompatible) and movement velocity (short time to peak force, long time to peak force) were administered in counterbalanced order across participants. The required response fingers always varied randomly within each block and were equiprobable. Recording The electroencephalogram (EEG) was recorded from F3, F4, C3 0,C4 0 (4 cm to the left and right of Cz, respectively), P3, P4, O1, and O2 with tin electrodes embedded in a nylon mesh cap (Electrode-Cap International; Eaton, OH), according to the International 10/20 system (Jasper, 1958). Left mastoid served as a reference. Horizontal electrooculograms (heog) were recorded from the left and right outer canthi and vertical electrooculograms (veog) from above and below the eyes. The EEG and EOG were amplified by SynAmps (Neuroscan, Inc.). Electrode impedances were below 5 ko. Band pass was set at 0.01 to 30 Hz ( 3dB). The electromyogram (EMG) was bipolarly recorded from the extensor digitorum muscle in the forearms with tin electrodes with a band pass of 13 Hz to 10 khz ( 3dB, Coulbourn Instrument V75-04). The EMG signals were full-wave rectified and integrated with a time constant of 100 ms (Coulbourn Instrument V76-23). All physiological signals, including the force curves, were digitized at a rate of 500 Hz. Data Analysis To obtain LRPs and other lateralized ERPs, signals from lateral EEG (F3/F4, C3 0 /C4 0, P3/P4, and O1/O2) and horizontal EOG electrodes were calculated to response hand specific bipolar derivations in the same way as for the LRP (Coles, 1989). That is, for each response, hand condition signals at ipsilateral recording sites were subtracted from the signals at homologous contralateral recording sites. Separate mean difference wave forms were computed for trials requiring left- and right-hand responses. These difference wave forms were averaged separately for each participant and each experimental condition. This subtraction procedure was employed to calculate both stimulus- and response-synchronized LRPs. Trials with incorrect responses were excluded from the analyses, as were all trials where vertical or horizontal EOG voltages exceeded a threshold of 100 mv during the recording epoch. To determine EMG onsets, we adhered to the procedure suggested by Smid, Mulder, Mulder, and Brands (1992). For each trial, the onset of the EMG response was determined by

5 224 H. Masaki et al. looking backward from the downward slope of the rectified- EMG peak for the time at which the amplitude was equal to or lower than the amplitude at the preceding sampling point. The validity of these onset measures was checked by visual inspection. We used this EMG onset as a trigger for both averaging of the EMG-synchronized LRP and for measuring EMG latency. The same algorithm was applied to the force curves to determine force onsets. Reaction time was measured as the interval between stimulus and response force onsets. LRP onset was determined as the intersection of the best fitting linear regression line through the slope of the LRP with the baseline, defined as the average voltage of the LRP during the 200 ms prior to stimulus onset (Mordkoff & Gianaros, 2000; Schwarzenau, Falkenstein, Hoormann, & Hohnsbein, 1998). The optimal regression line was selected as follows: First an anchor point was chosen; for the EMG-synchronized LRP, one end of the regression line was anchored at the point of peak amplitude; for the stimulus-synchronized LRP, it was anchored at the point on the LRP associated with the mean EMG onset. We did not anchor the last point of the regression line at the peak of the stimulus-synchronized LRP wave as suggested by Mordkoff and Gianaros (2000) because in the slow time-topeak-force condition the LRP did not show a clear peak. Then, series of least squares linear regressions were calculated based on the anchor point and a specified number of data points preceding the anchor points. Regression lines moved freely along the y-axis also at the anchor point. Starting with n 5 2, the number of data points preceding the anchor points was incremented in steps of one until the variance accounted for by the n 1 1st line was less than for the nth line and the algorithm was stopped. At this time, the nth regression line was visually inspected to determine if it was only a local maximum (i.e., not the true best fit). If the nth regression line appeared to be an obviously poor fit, then the algorithm was restarted until the true optimal regression line was determined (maximum variance accounted for). The point at which this optimal regression line intersected with the baseline was taken as the LRP onset. The regression lines were calculated in jackknife averages, according to the procedure suggested by Miller, Patterson, and Ulrich (1988). When subjecting the onset measures from jackknife averages to analyses of variance (ANOVA), F values were corrected as Fc 5 F/(n 1) 2 (Ulrich & Miller, 2001). Results Performance Table 1 shows mean RTs and mean EMG latencies of correct responses. S-R compatibility and time to peak force seem to Table 1. Means and SEM (Milliseconds) of EMG Latency (Stimulus-to-EMG Onset) and Reaction Time (Stimulus-to-Force Onset) as a Function of Compatibility and Movement Velocity EMG latency Reaction time Compatible Fast (17.0) (17.7) Slow (22.9) (21.6) Incompatible Fast (31.5) (29.2) Slow (31.1) (26.6) Table 2. Means and SEM of Hand-Choice Errors (%), Incorrect Response Force and Incorrect Time-to-Peak-Force as a Function of Compatibility and Movement Velocity Hand-choice errors Peak-force errors Time-to-peak-force errors Compatible Fast 0.63 (0.33) 1.67 (0.55) 8.68 (2.05) Slow 0.76 (0.42) 2.36 (0.67) (2.33) Incompatible Fast 3.68 (0.78) 2.71 (0.94) (2.53) Slow 3.95 (0.86) 2.85 (0.51) (2.29) exhibit additive effects on both dependent variables, suggesting that these two factors selectively affected different processing stages. ANOVA on mean RTrevealed a significant main effect of S-R compatibility, F(1,11) , po.001, but no effect of time to peak force, F(1,11) , p There was no interaction between these variables, F(1,11) , p Mean EMG latency, however, yielded significant main effects of both S-R compatibility and time-to-peak-force manipulations, F(1,11) , po.005; F(1,11) , po.005, respectively. Importantly, there was no interaction between these factors, F , confirming independent effects on separate processing stages. Table 2 shows percentages of hand-choice errors and responses failing to meet the time-to-peak-force and force criteria (force errors). For hand-choice errors, ANOVA indicated that incompatible trials induced more hand errors than compatible trials, F(1,11) , po.001. Thus, speed accuracy trade-off did not occur. No reliable effects of the experimental variables were seen for the force errors and time-to-peak-force errors. We should like to point out that these latter types of errors do not relate to the integrity of the transmitted hand information, but merely reflect imperfections of movement execution. Therefore, these errors do not call into question the applicability of the additive factors logic in the present case as an excess of choice errors would. Stimulus-Synchronized LRPs Figure 1 (left panel) shows the grand averaged wave forms of the stimulus-synchronized LRPs over central regions (C3 0 /C4 0 ), other stimulus-synchronized lateralized wave forms, and the lateralized horizontal EOG. As can be seen in this figure, the S-R compatibility manipulation seemed to affect the latencies of the stimulus-synchronized LRPs. Figure 2 (left panel) shows a magnification of the stimulus-synchronized LRPs over central regions with the averaged best-fit regression lines in each condition superimposed. Table 3 presents the mean onsets of the stimulus-synchronized LRP measured with the regression line method in the jackknife grand averages. ANOVA revealed a significant S-R compatibility effect, Fc(1,11) , po.05, showing earlier onsets in the compatible than in the incompatible conditions. Otherwise neither main effect of time to peak force, Fc , nor interaction with S-R compatibility, Fc(1,11) , p4.10, was found. Furthermore, comparison of the slopes of the regression lines through the initial flank of the stimulus-synchronized LRP revealed an S-R compatibility effect, Fc(1,11) , po.01, indicating steeper slopes in compatible than incompatible conditions (Table 4). However, an interaction between variables

6 Functional locus of the LRP 225 Figure 1. Left panel: grand-average stimulus-synchronized lateralized readiness potentials (LRPs) at four homologous electrode positions, for the lateralized horizontal electrooculogram (heog), rectified EMG (remg), and force waveform. The vertical line indicates stimulus onset. C/F: fast compatible condition; C/S: slow compatible; IC/F: fast incompatible; IC/S: slow incompatible. Right panel: grand-average EMG-synchronized LRPs, heog, remg, and force waveform. The vertical line indicates the EMG onset. was also significant, Fc(1,11) , po.05. Considering Table 4, it is clear that this interaction stems from the relatively steep slopes in the fast compatible condition; Bonferoni-corrected pairwise comparisons indicated that this condition differed significantly from both incompatible conditions and as a trend from the slow compatible condition, p Figure 1 (left panel) also indicates a small negative-going parieto-central deflection in the LRP in the compatible conditions about 200 ms after the stimulus. This deflection was absent in the incompatible conditions. This effect was analyzed by measuring mean amplitudes between 150 and 250 ms. ANOVA revealed a significant interaction of S-R Compatibility Electrode, F(2,22) , po.05, indicating a strong trend for a larger negativity in the compatible than the incompatible condition at parietal sites, t(11) , p The reader will note in Figure 1 (left panel) that there was another negative deflection around 350 ms after stimulus onset at parietal sites. Because in this time range there is also strong central activity, it seems to be most sensible to assume that the parietal activity reflects volume-conducted LRP. Response-Synchronized LRPs EMG onsets were used for calculating response-synchronized LRPs (Figures 1 and 2, right panels). Clearly, there are effects of the experimental manipulations on the onsets and the negativegoing flanks of the LRP and also on the time course following EMG onset. Figure 1 (right panel) also suggests that both the force curves and the rectified EMG wave forms were not affected by the S-R compatibility manipulation, whereas both showed greater peak latencies in the slow movement relative to the fast movement, reflecting the task requirements. It appears from the figures that, especially in the fast compatible condition, the LRP may start very early (o 300 ms), increasing at first slowly until about 150 ms and then steeply. Apart from being difficult to explain, such a biphasic time course might call into question the applicability of the linear

7 226 H. Masaki et al. Figure 2. Left panel: grand-average stimulus-synchronized LRPs at central regions with average best-fit regression lines for onset determination superimposed. Stimulus onset is at time zero. Right panel: grand-average EMG-synchronized LRPs with best-fit regression lines. The vertical line indicates EMG onset. regression method. Therefore, we first assessed the onsets of the LRP wave shapes in each of the four experimental conditions by administering serial one-tailed t tests against zero at each digitized time point for the C3 0 /C4 0 electrode pair (e.g., De Jong et al., 1988; Smid et al., 1987; Valle-Inclán & Redondo, 1998). These tests indicated relatively late and very similar significant LRP onsets for the fast compatible and fast incompatible conditions ( 140 and 142 ms, respectively); markedly earlier in onset were the slow compatible and slow incompatible conditions ( 218 and 192 ms, respectively). In addition, ANOVA on the average amplitudes of the 300 to 200 ms time segment including frontal to parietal electrode pairs did not yield any significant effects. These results indicate that the very early deviations ( 400 to 150 ms) from baseline to be seen in the fast and to some extent Table 3. Mean Onset (Milliseconds) and SEM of the Stimulus- Synchronized LRP Relative to the Stimulus and Mean Onset (Milliseconds) of the Response-Synchronized LRP Relative to the EMG Onset Stimulus to LRP interval LRP to response interval Compatible Fast (1.18) (1.66) Slow (1.58) (5.14) Incompatible Fast (1.30) (1.84) Slow (2.20) (1.77) Table 4. Mean Slope (Microvolts per Millisecond) and SEM of Regression Lines Applied to the Flanks of the Stimulus- Synchronized LRP and the Response-Synchronized LRP a Stimulus-synchronized LRP Response-synchronized LRP Compatible Fast (0.0003) (0.0003) Slow (0.0002) (0.0003) Incompatible Fast (0.0003) (0.0003) Slow (0.0002) (0.0003) a More negative values represent steeper slopes. also in the slow compatible condition are not significant, which justifies the use of the regression method for statistical analyses of condition effects. In addition, the numerical pattern of the onsets suggests a sizeable effect of movement velocity, whereas compatibility is much less effective. The small negative-going parietocentral deflection in the LRP in the compatible conditions was also analyzed by measuring mean amplitudes between 300 and 200 ms before the EMG onset. In contrast to the stimulussynchronized LRP, ANOVA showed neither main effect of S-R compatibility nor interaction of S-R compatibility and electrode, Fs(1,11) and 1.99, ps 5.17 and.16, respectively, for this deflection. Table 3 shows the onsets of the grand averaged EMGsynchronized LRPs as measured with the regression method. ANOVA of the onsets thus obtained revealed a significant timeto-peak-force effect, Fc(1,11) , po.05, confirming longer intervals between the LRP and the EMG onsets in the slow relative to the fast time-to-peak-force condition. Neither S-R compatibility effect, Fc , nor interaction between variables, Fc , was observed. ANOVA of the slopes of the regression lines (Table 4) revealed a time-to-peak-force effect, Fc(1,11) , po.01, supporting the observation in the curves (Figure 2, right panel) that the LRP shows a steeper rising flank in the fast relative to the slow time-to-peak-force condition. No S-R compatibility effect on the slopes was present, Fc However, an interaction of S-R Compatibility Time to Peak Force was revealed, Fc(1,11) , po.05, suggesting that the time-to-peak-force effect in the slopes is even more pronounced in the compatible condition. Bonferoni-corrected pairwise post hoc comparisons indicated that the compatible slow condition differed significantly from the compatible fast condition, po.01) andfas a trendffrom the incompatible fast condition, p 5.08); in addition there was also a trend for a difference between the compatible fast and the incompatible slow conditions, p Another interesting observation concerns the time course of the response-synchronized LRP amplitude. Whereas immediately prior to the response there appear to be no pronounced amplitude differences, the LRP after EMG onset was flatter in the slow than in the fast time-to-peak-force conditions. Thus LRP amplitude is largest in the fast condition until about 150 ms

8 Functional locus of the LRP 227 after EMG onset but smallest thereafter. These observations were supported by ANOVAs of mean LRP amplitudes, measured at the central electrodes in four consecutive 100-ms time segments starting at EMG onset. A main effect of time to peak force appeared in the first segment after EMG onset and in the final ( ms) segment, F(1,11) and 43.74, po.05 and o.001, respectively, but not in between, F Interestingly, in the time segment immediately prior to EMG onset, LRP amplitude was significantly larger in the compatible than in the incompatible conditions (M vs mv), F(1,11) , p.05. The effect was present also when the measurement segment was narrowed to 30 to 40 ms prior to EMG onset, F(1,11) , po.05. Importantly, there was no effect of time to peak force on the preresponse LRP amplitude (100-ms segment before EMG onset: F ; 30 to 40 ms before EMG onset: F , respectively). Discussion It was the aim of the present study to achieve a more precise locus of the onset of the LRP within the information processing system. To this aim we factorially manipulated stimulus response compatibility and movement velocity, taken to selectively influence the response selection and motor programming stages, respectively. EMG latency as measured in the EMG clearly showed additive effects of these variables without any interaction, indicating that separate stages had been manipulated selectively (Sternberg, 1969). Thus, the findings of Spijkers (1989) that movement speed affects motor programming are extended here to isometric conditions. The results for RT as measured in force onsets were similar to those for EMG latency as measured in EMG onsets but were less clear, most likely due to the problems noted above in determining the onsets of the force curves in the long time-to-peak-force conditions. A further important requirement for the validity of the present design is that the task is accomplished within a single processing route and does not involve additional response activation via a second, unconditional route. The absence of incorrect LRP activation at central electrodes in the incompatible conditions indicates that this was indeed the case. Together, these findings make us feel confident that the present design did fulfill the important requirement of independent manipulations of the time demands for response hand selection and motor programming in a serial processing chain. The effects of the experimental manipulations on the LRP onsets were quite clear cut. The S-R compatibility manipulation exclusively acted on the stimulus-lrp interval and had no effect on the LRP-response interval. Conversely, movement velocity had no effect on the stimulus-lrp interval but did affect the LRP-response interval in the predicted direction. As outlined in the introduction, this pattern is expected when the LRP starts after the completion of response hand selection and at the beginning of motor programming. A problem for this interpretation may be seen to arise from the effect of compatibility on the slope of the stimulus-synchronized LRPs in the fast movement condition, which would be in line with LRP elicitation during response selection. If this was the case, however, S-R compatibility should also affect the slopes of the response-synchronized LRPs, which it did not. Inspection of Figure 2 and Table 4 shows that the compatibility effect on the slope of the stimulus-synchronized LRP is due to the flattening of the slope in the fast incompatible relative to the fast compatible condition. We suggest that this is due to an increase of variability in the EMG latency specifically in the fast incompatible condition (cf. Table 1), which would smear the averaged wave shape and thus flatten the slope. In line with this interpretation, the slopes and amplitudes are very similar for these conditions in the responsesynchronized LRPs, where the jitter due to RT variability is eliminated. LRP elicitation during response hand selection would also induce an S-R compatibility effect in response-synchronized onsets, which was neither indicated by the serial t tests nor by the regression method. Note that although S-R compatibility appears to influence response-synchronized LRP onset in the slow condition (cf. Table 3), this effect not only fails to achieve statistical significance, it is also opposite in direction to what LRP-elicitation during response hand selection would predict: Rather than being shorter in the compatible as compared to the incompatible condition, the LRP-response interval is actually somewhat longer. From the present findings we can also rule out that the LRP starts at some point during motor programming. In that case, LRP onset relative to the stimulus should be delayed when the time demands for motor programming are increased. This was not the case, however; movement velocity had no effects on stimulus-synchronized LRP onset. Our data also rule out the two other theoretically possible alternatives, beginning of LRP (1) prior to or at the beginning of response hand selectionfin which case one would not expect any experimental effects on the stimulus-synchronized LRP onsetfand (2) after motor programming, as indicated by an absence of experimental effects on the response-synchronized LRP onset. Both predictions are at odds with our data. Therefore, we conclude from the present findings that the onset of the LRP follows response hand selection and starts with the beginning of motor programming. This conclusion is consistent with most of the other available findings. Thus the stimulus-lrp interval is affected by stimulus quality (Smulders, 1993; Smulders et al., 1995) and symbolic compatibility (Smulders, 1993). The LRP-response interval is affected by response-complexity (Smulders et al., 1995), by preliminary information about response parameters (Leuthold et al., 1996), and by the compatibility (Miller & Ulrich, 1998) and number of alternatives (Hackley & Valle-Inclán, 1999) of finger choice at constant levels of hand selection difficulty. The only inconsistency that we can see is the observation of Smulders (1993) that the LRP-response interval is delayed for degraded incompatible stimuli. Are the present findings called into question by the small early parieto-central negative dip in the stimulus-synchronized LRPs in the compatible conditions? Such dips have been reported when lateralized stimulus presentation is confounded with response side (cf. Eimer, 1996) and have been interpreted as indicating automatic response activation by stimulus features that overlap with a response dimension. One might argue that such automatic response tendencies were also elicited by our centrally presented letter stimuli. In this case, we can rule out two possible processes. First, the parietal dip is not an LRP proper because it is larger over parietal than over central sites whereas the LRP proper shows the opposite distribution (cf. Figure 1). Therefore the parietal dip does not reflect immediate motor activation. Second, the dip is also not an N2pc as observed in visuospatial attention (e.g., Luck & Hillyard, 1994; Praamstra & Oostenfeld, 2003; Wascher & Wauschkuhn, 1996) because this component is typically observed with lateralized horizontal stimulus arrays and shows an occipitotemporal scalp distribution. On the other hand, Stu rmer and Leuthold (2003) have recently reported an

9 228 H. Masaki et al. early parietal lateralized ERP component for vertical stimulus arrays. This component could be functionally dissociated from the LRP proper and was suggested to represent externally triggered transformations of spatial stimulus features into spatial response features. Possibly, the letter stimuli of the present study might have triggered spatial response features in a similar vein as did the spatial stimulus features of Stu rmer and Leuthold. The early parietal dip is not necessarily at odds with a single route account of the present paradigm (Kornblum et al., 1990); possibly, the parietal activation represents a processing stage preceding the motor activation reflected in the LRP proper. Because previous studies using symbolic stimuli have rarely recorded lateralized ERPs from electrodes other than central ones, the generality of such parietal activation is hard to determine. Therefore, without further research it cannot be ruled out that the parietal activation takes place in a route that runs parallel to the conditional response selection. Independent of the functional interpretation of the parietal dip, our findings were not affected by its presence because of its small size at central electrodes and because we used the regression method for onset determination, which is robust against small deviations from the baseline. Together, we suggest that the LRP starts as soon as there is complete determination of the response hand by the response selection stage. The response hand code appears to be transmitted to the motor stages in a discrete fashion; if partial information about the response hand were transmitted, for example, while still being determined, we should have observed an effect of stimulus response compatibility also on the responselocked LRP, which was not the case. How can these conclusions be reconciled with the abundant evidence about partial transmission of information and about effects of the number of stimulus alternatives and compatibility on the response-locked LRP? First of all, let us clearly state that the present findings were obtained in a condition where the stimulus contained only a single attribute (letter shape) and where, according to common understanding, there was only a single processing route. Evidence for partial transmission of information into the motor system comes from experiments with multiple stimuli, such as the Eriksen flanker task (Gratton et al., 1998), or with stimuli having several attributes, as used in two-choice go/no-go tasks (e.g., Abdel-Rahman, Sommer, & Schweinberger, 2002; Miller & Hackley, 1992; Osman et al., 1992) or in the Simon task (e.g., Stürmer et al., 2002; Valle-Inclán, 1996; Valle-Inclán & Redondo, 1998). In all these designs, a biphasic or temporary LRP activation may occur. For example, early activation of the incorrect response followed by activation of the correct response is seen for incompatible flankers in the Eriksen task and for noncorresponding stimuli in the Simon task. Temporary response activation is seen in the LRP for no-go stimuli in twochoice go/no-go tasks. According to the present findings, these signs of temporary response activations should all emerge after response hand selection at the beginning of motor programming. Such an explanation fits very well with dual route models as have been proposed to explain the Simon effect or spatial S-R compatibility (Kornblum et al., 1990). In this case, the motor programming stage would receive two volleys of activation, one via the conditional route, involving controlled response selection, and another one via a direct link from perceptual processes, bypassing response selection. Temporary LRP activations in other designs with multiple stimuli or multiple stimulus attributes might be explained similarly with multiple streams of activation reaching the motor programming stage. These streams of activation might be caused by multiple response selections as in dual task designs (e.g., Osman & Moore, 1993; Sommer et al., 2001) or by additional links from processes other than response selection (Hommel, Mu sseler, Aschersleben, & Prinz, 2001). Studies that have shown effects of stimulus response compatibility (Miller & Ulrich, 1998) or number of response alternatives (Hackley & Valle-Inclán, 1999) on the responsesynchronized LRP indicate that response selection itself may consist of several aspects, only one of them being the selection of the response hand. When other aspects in addition to hand, such as response finger, have to be selected as well, these processes appear to follow hand selection at least in the existing studies because they affect the interval after LRP onset. If this interpretation is correct, it also follows that different codes derived during response selection can be transmitted separately. As far as the motor programs are concerned that are presumably assembled and implemented after response selection, it is not completely clear whether the effector-specific programming phase is preceded by a more abstract phase that is independent of the specific muscle groups required for the response (Schmidt, 1975). According to suggestions by Ulrich et al. (1998) and dipole models by Leuthold and Jentzsch (2001, 2002), the LRP reflects effector-specific motor preparation taking place within the premotor area and primary motor cortex. If this holds true, the assembling and selection of abstract movement programs within the supplementary and cingulate motor areas would take place before LRP onset. Future research should test this notion. In this context, it is also interesting to consider variables that affect LRP amplitude. As far as movement kinetics are concerned, the data are somewhat controversial, as outlined in the introduction. Whereas Sommer et al. (1994) and Mordkoff and Grosjean (2001) did not find effects of response force and time to peak force, Ray et al. (2000) reported larger LRPs for higher force growth rates. Above, we hypothesized that these discrepancies may relate to requirements for on-line control over nonballistic movements. In the present experiment, the fast movement with a time to peak force of ms is at the lower limit and therefore is most probably not under control while it is executed. In contrast, the slow movement with a time to peak force of ms not only requires more programming prior to execution (as confirmed by present LRP onset data) but also renders itself to on-line control. Interestingly, we did find an effect of time to peak force on LRP amplitude, but only after response onset. Following EMG onset, LRP amplitude in the slow condition showed a relatively steady activation over time, whereas in the fast condition, the typical phasic activation with a peak around 100 ms appeared. These difference in the time course of the LRPs following response onset may relate either to differential needs for on-line response regulation (Van Donkelaar & Franks, 1991) or in differential reafferent input from cutaneous and muscle receptors. Prior to response onset, however, there was no effect of time to peak force on LRP amplitude, which is in line with the reports of Sommer et al. (1994) and Mordkoff and Grosjean (2001) and rules out an explanation of the results of Ray et al. (2000) in terms of differential preparation for on-line control. In conclusion, manipulating response hand selection and motor programming in a factorial design, we obtained evidence that further specifies the locus of the LRP onset within the information processing system. According to these findings the LRP starts (1) after the completion of response-hand decision and (2) at the beginning of motor programming. If these