Recording mobile EEG in an outdoor environment reveals cognitive-motor interference dependent on movement complexity

Oftentimes we find ourselves in situations in which we need to perform concurrent motor and cognitive tasks like simple locomotion while being cognitively involved. In the present study, we investigated in how far cognitive and motor functioning interfere in an outdoor environment. Our participants performed an auditory oddball task while concurrently completing various motor tasks on the outside premises of our institute. Beside behavioural responses and subjective workload ratings, we also analysed electrophysiological data recorded with a 30-channel mobile EEG montage. We observed an increase of subjective workload and decrease of performance with increasing movement complexity. Accordingly, we also found a decrease in the parietal P3 amplitude as well as in frontal midline Theta power with higher motor load. These results indicate that an increased movement complexity imposes a higher workload to the cognitive system, which, in turn, effectively reduces the availability of cognitive resources for the cognitive task. Overall this experiment demonstrates the feasibility of transferring classical paradigms of cognitive research to real-world settings. The findings support the notion of shared resources for motor and cognitive functions by demonstrating distinct modulations of correlates of cognitive processes across different motor tasks.

measures, as we analysed the EEG time-locked to the onset of the stimuli of the cognitive task. The amplitudes of stimulus-locked P2, N2 and P3 components as well as frontal Theta power should be decreased with higher movement complexity, as fewer attentional resources are available to the cognitive task. According to recent findings, target probability should affect response times. As a proof-of-concept, we correspondingly expect slower response times with lower target probability 51 . Also P2, N2 and P3 amplitudes within targets should be diminished as target probability is increased [34][35][36] , as should Theta power 52 .
Results nASA-tLX. All dimensions on the NASA-TLX are poled and analysed with low rating values reflecting low subjective workload. Figure 1 illustrates the average ratings and the standard error of all subscales of the NASA-TLX for each experimental condition combination. All results of the separate ANOVAs are shown in Table 1.
Cognitive demand. The target probability main effect showed that participants rated high target probability (M = 3.6, SD = 1.99) as significantly more cognitively demanding (F 1,19 = 6.73, p = 0.02, p crit = 0.033, η p 2 = 0.26) than low target probability (M = 3.1, SD = 1.64). Regarding the main factor of movement complexity, more complex movement was linked to the higher cognitive demand rating (M stand   p < 0.001, p crit = 0.05), though ratings of walking and the obstacle course did not deviate significantly after correction (t 38 = −2.14, p = 0.04, p crit = 0.017). There was neither an influence of target probability on the rating of physical workload, nor a significant interaction term.
Temporal demand. Concerning the temporal demand, only the main effect of movement complexity showed an acceptable effect size -but no significance (F 2,37,96 = 2.95, p = 0.06, p crit = 0.05, p 2 η = 0.13) -with only walking and the obstacle course displaying a significant difference (t 38 = −2.42, p = 0.02, p crit = 0.05). Interestingly, ratings for the obstacle course were the highest (M = 5.50, SD = 2.53), followed by standing (M = 4.90, SD = 2.63) and then walking (M = 4.48, SD = 1.77) regarding simple means. Therefore, the hypothesized change in difficulty regarding movement complexity did not translate into higher ratings. Similar to the previous measure, neither target probability nor the interaction of both main effects revealed significant differences.
Frustration level. There were no differences regarding the subjective frustration rating, neither for both main effects, nor for the interaction term.     eRps. Frontal N2 at Fz. The frontal N2 ERP component did not vary significantly with target stimulus or movement complexity condition. Significant amplitude differences could be found for the factor target probability  Figure 4 shows ERPs at Pz, topographies at the given time windows, and the interaction plots of parieto-temporal P2, N2, and P3 amplitudes.  The participant wore a 30-electrode cap, which was connected to a mobile amplifier placed in a pocket at the back of the head. In the backpack a raspberry Pi was stored which generated auditory stimuli and trigger signals. Triggers were transferred into digital signals via a Trigger extension box which then was connected to the amplifier. The participant held a handle with a reaction button in the right hand. By pressing the button with the right thumb, response triggers were inserted into the EEG-signal via the Trigger extension. EEG-data was stored offline on an SD-card inside the amplifier. www.nature.com/scientificreports www.nature.com/scientificreports/ SD = 1.00) compared to walking (M = 0.90, SD = 0.83; t 38 = 4.98, p < 0.001, p crit = 0.033) and the obstacle course (M = 0.88,SD = 0.78; t 38 = 5.12, p < 0.001, p crit = 0.05). The main effect of target stimulus was also significant (F 1,19 = 27.91, p < 0.001, p crit = 0.05, η p 2 = 0.59), with target stimuli eliciting larger Theta responses than standard stimuli. A significant interaction was found for the factors target probability and target stimulus (F 1,19 = 9.99, p = 0.005, p crit = 0.036, η p 2 = 0.34). By looking at specific interaction effects within target probability factor levels, further post-hoc pairwise comparisons demonstrated that the difference between target (M = 1.29, SD = 0.91) and standard stimuli (M = 0.89, SD = 0.76) was larger in LTP blocks (t 38 = −6.14, p < 0.001, p crit = 0.05) compared to target (M = 1.56, SD = 1.01) and standard stimuli (M = 0.83, SD = 0.68) in HTP blocks (t 38 = −3.35, p = 0.002, p crit = 0.025). Frontal ERSPs of Theta power at Fz, interaction plots of frontal theta power, and time-frequency plots at Fz are depicted in Fig. 5.

Discussion
In this study, we used a mobile and lightweight EEG measurement setup and recorded subjective, behavioural, and electrophysiological measures of participants performing a cognitive-motor dual-task in an outdoors environment. While performing an auditory oddball task, the participants either stood, walked laps, or completed the laps with obstacle course elements. In contrast to previous studies on cognitive-motor interference dual-tasks 1,14,53 , we pursued an approach to investigate basic attentional mechanisms embedded in ecologically valid conditions. Recent advances in the technological development of recording equipment 9 and stimulus presentation 54 increased the feasibility of mobile experiments. Also, prior studies using mobile EEG approaches outside of the laboratory demonstrated that high ecological validity does not necessarily decrease signal quality [11][12][13]55 .
Since the data of all experimental conditions were collected at the same location, we kept environmental factors like temperature, lighting or sensory complexity comparable.
With respect to the manipulation of movement complexity the results are clearly in line with our initial hypothesis. The behavioural measures indicate that a higher complexity of motor activity is accompanied by a decline in performance. Response latencies were significantly increased in the walking and the obstacle course condition as compared to the standing condition. There was also a tendency towards slower responses in the obstacle course compared to the walking condition. A similar pattern emerges for omission errors, since during the obstacle course participants missed significantly more targets than during the other movement complexity conditions. Keeping in mind that the motor task should be prioritized by the participants in order to reduce the risk of injury, this overall decrement in performance may indicate a decreased availability of attentional resources for cognitive performance with higher motor workload. This interpretation is also supported by the subjective ratings, as the participants evaluated the cognitive and physical demand, as well as general effort higher and also rated their own performance lower with increasing complexity of the motor conditions. This cognitive-motor interference is clearly visible in the EEG as well. As hypothesized, the parietal P3 amplitude demonstrated a significant distinction between all movement complexity conditions. The P3 amplitude in response to target stimuli was significantly decreased in both locomotion conditions as compared to the standing condition. As the parietal P3 has been linked to the allocation of cognitive resources 38,44 , this observation further strengthens the interpretation of the reduced availability of attentional resources to be responsible for the decline of performance with increasing movement complexity. The analysis of frontal midline Theta power revealed comparable results. The event-related increase in Theta power was significantly decreased in the walking and the obstacle course condition compared to standing, indicating a decline in focused attention for mental task execution while in motion 47,56 . Also, it could be shown, that in conditions with higher frontal Theta power (standing), participants responded fastest. By this connection Theta power might reflect cognitive effort and could therefore allow for a connection between electrophysiology and resource allocation 46 . Also, when referring to the review of Roux and Uhlhaas 57 about oscillations and working memory, there is a clear connection between working memory maintenance and Theta activation. Decreased Theta power during higher motor load could therefore indicate impaired working memory maintenance processes and account for the deteriorated response times. Altogether, regarding prior dual-task interference literature, these electrophysiological results suggest that both the motor and the cognitive task share a common pool of attentional resources 31,58 . This is in line with previous findings, demonstrating differences in resource allocation for non-motor and motor tasks 1,13,14,59 .
Interestingly, when looking only at standard stimuli, P3 amplitude was significantly negatively enhanced only for walking, possibly indicating cognitive processing of no-go stimuli. Our participants also exhibited a significantly smaller number of omission errors while walking as compared to the standing and obstacle course conditions. These results may in theory indicate a non-linear variation of task-difficulty from standing to walking to obstacle course, at least with respect to specific task-demands. The analysis of the gait-measures, however, does not support this interpretation. Participants took more steps and had a higher variability between steps during the obstacle course compared to normal walking, but completed a similar number of laps. Considering the review by Al-Yahya et al. 25 these results might indicate a difference of overall workload induced by the movement complexity, especially since the obstacle elements forced a certain gait-pattern onto the participants, so that a significant difference in workload might be caused by correct task-execution itself.
The analysis of the experimental factor target probability partially suggested the replication of previous laboratory findings. Being in line with our hypothesis, subjective ratings indicated less cognitive effort and a better subjective performance for the LTP condition. Adversely, response times were slower for LTP stimuli, while participants made less omission errors. In contrast to our expectations as well as to previous findings 36,52 , the frontal Theta power and the parietal P3 amplitude were not modulated by the factor target probability. The N2 amplitude, however, revealed a main effect of target probability with larger negative deflections in the LTP compared to the HTP condition. Target-probability effects for the N2 have rarely been researched and adverse findings have been reported from the visual 34 www.nature.com/scientificreports www.nature.com/scientificreports/ reflect a greater need for cognitive control 41,62 when there are less targets, which might also explain slower responses. We also observed a significant interaction of the factors target stimulus and target probability for frontal Theta power, reflecting higher Theta power for targets within LTP compared to HTP as hypothesized 52 . In summary, participants rated themselves to be better, had more negative N2 amplitudes and higher Theta power, but showed slower response times during the LTP conditions. It thus seems that a higher target probability does not influence cognitive workload, but seems to influence the cognitive system in a rather general manner. Thus, certain effects of target probability found in a laboratory setting were present in this study, main findings of target P3 amplitude variation could not be replicated.
Overall, the results clearly show that executing a well-known laboratory paradigm in a natural environment is feasible, which is in line with previous findings 12,14,55,59 . Besides replicating the oddball effect outdoors, the data provides new evidence on the topic of cognitive-motor interference. Firstly, a clear limitation of the presented approach is that the task is very basic and therefore lacks ecological validity. Secondly, presenting auditory stimuli over passive noise-cancelling headphones might induce additional cognitive demand, since perception of naturally occurring environmental stimuli and, consequentially, orientation in said natural environment in can be impaired. Due to the jittered stimulus presentation these additional cognitive demands should not induce systematic error variance though. Finally, to keep results comparable to prior literature, we ensured that external influences (temperature, lighting etc.) were similar during each conduction of the experiment which might restrict ecological validity further. Future research should address this limitation by investigating cognition in more complex and application-related scenarios. Nevertheless, it could be shown that a manipulation of movement complexity led to distinct effects in subjective, behavioural and electrophysiological measures. The fact that motor complexity and not speed was manipulated provides strong evidence that an increased motor load reduces the cognitive resources available for cognitive processes.

participants.
In this study 29 healthy subjects participated. All of the participants were free of prior or present neurologic or psychiatric conditions, right-handed, and had normal or corrected-to-normal vision. None of the participants reported hearing deficiencies. Due to problems concerning trigger generation in early stages of the experiment, only datasets of 20 participants (11 male, 9 female) could be used for further analysis. The subjects' age in the concluding sample ranged from 20 to 30 years (M = 24.10, SD = 3.08). Subjects were paid 10 € per hour and gave their informed consent. The study was approved by the local ethics committee of the Leibniz Research Centre for Working Environment and Human Factors and was conducted in accordance with the Declaration of Helsinki.
Apparatus and stimuli. The whole data acquisition took place on the outside terrain of the institute between June 26 th and August 29 th 2018. The obstacle course used for implementing the most complex motor task consisted of two small staircases, two balancing beams, and two boards with punched-out holes (Fig. 6c). The course for the walking and the obstacle course condition had a circumference of 75 meters. The obstacle course was situated on the lawn on the outside premises of the institute, so subjects walked on a grass field with common surface irregularities during the walking or obstacle course conditions. The outside temperatures at the beginning of the experimental procedure ranged from 15 °C to 28 °C.
To be able to record EEG data in this natural setting, a mobile setup for stimulus generation and data acquisition was used. A Raspberry Pi 2B (Raspberry Foundation, UK) with custom OpenSesame Scripts 63 was used for stimulus presentation and -in combination with the LiveAmp Sensor and Trigger Extension (Brain Products GmbH, Gilching, GER) -for trigger generation. To enable behavioural responses by the subject, a self-made handle with a response button was used which was connected to the Trigger Extension. The subjects were instructed to operate the handle with their right hand (Fig. 6b). The devices were stored in a small backpack which the participant wore for the duration of the experiment. The picture shows the first author of this manuscript during a pilot study. Written informed consent to publish this picture was obtained. In order to prevent induction currents due to cable sway 64 , all wires were prepared not to cross each other and were taped together for protection from movement.
Auditory stimuli were presented to the participants via passive noise-cancelling in-ear headphones (Bose QC25, Framingham, USA). The tones presented consisted of pure sine waves (low: 600 Hz, high: 900 Hz) with a length of 62 ms (10 ms rise/fall time). Stimuli were presented with an inter-stimulus interval of 2000 ms with a jitter of +/−250 ms. procedure. Upon arrival between 8 and 9 am, the participants read the experiment information and signed the informed consent. Subsequently, they were fitted with a 30 electrode actiCap setup (actiCAP Slim electrodes & actiCAP Snap cap, Brain Products GmbH, Gilching, GER) as well as the backpack and were accompanied outside to the obstacle course track. For each participant the preparation took approximately 1 hour.
At the course the participants were fully instructed about the task execution. For the dual-task approach participants had to perform a cognitive and a motor task simultaneously. The cognitive task consisted of a simple signal-detection auditory oddball task. Participants were either presented with a low-pitched standard-tone (600 Hz) at an average sound level of approximately 65 dB(A) or with a high-pitched target-tone (900 Hz) at an average sound level of approximately 72 dB(A) in a randomized stream of stimuli. Sound levels were measured directly at the earphones' earmold sound outlet. The participants were instructed to respond to a target tone by pressing the response button and to withhold a response when hearing a standard tone. The stream of stimuli presented to the participants consisted either of 80% standard and 20% target tones in the LTP condition or of 65% standard-and 35% target-tones in the HTP condition. For the motor task, participants had to either stand still on a predetermined spot (stand) in the lower left corner of the obstacle course, walk around the obstacle course (walk), or include the obstacles of the course while walking (obstacle course, see Fig. 6a). Whether a participant performed the motor task clockwise or anticlockwise was quasi-randomized and counterbalanced across participants. The sequence of tasks was quasi-randomized using a six factorial latin square design.
After each task, the participants' subjective task-load was assessed via the NASA-TLX 37,65 questionnaire. The participants also had the possibility for a short rest. For each factor combination of the 2 (target probability) × 3 (movement complexity) design the subjects were presented with 450 Stimuli. Each experimental block had a duration of 15 minutes, totalling to an overall duration of 1.5 hours. After finishing the task, the subjects were escorted back into the laboratory, where the electrode cap and the backpack were removed. Overall, the experimental procedure took roughly 3.5 to 4 hours per participant. electrophysiological data acquisition. Since mobile and real-world EEG setups are quite new, this section will explain the EEG equipment and subject preparation in higher detail than usual 66 . EEG data was acquired via 30 active electrodes in a standard 10-20-system montage (Fp1, Fp2, F3, F4, F7, F8, Fz, FC1, FC2, FC5, FC6, C3, C4, Cz, T7, T8, CP1, CP2, CP5, CP6, P3, P4, P7, P8, Pz, PO9, PO10, O1, O2, Oz). To fully prepare a participant for recording, a tight fitting, flexible cap with electrode holders was pulled over the subject's head at first. Sizes of caps ranged from 54 to 60 cm in circumference. Then, actively shielded electrodes were inserted into the electrode holders and filled with conductive gel until they reached an impedance of 10 kΩ or below. Because of the low weight, profile, and centre of gravity of the electrodes, mechanical electrode displacement was prevented to a considerable degree. Additionally, electrode cables were carefully aligned to not cross each other or sway around while the participant was moving. Therefore, all cables were routed through specific cable mounts of the actiCap next to the participant's ears. To further reduce cable motion, we bundled cables from the left and the right side and taped them together respectively. These cables were then routed to the mobile amplifier which was placed in a pocket at the back of the participant's head. Data were recorded with a LiveAmp amplifier (Brain Products GmbH, Gilching, GER) with a sampling frequency of 500 Hz and a bit depth of 24 bit. All data were stored on a micro SD card inside the amplifier while the online signal could be viewed via a Bluetooth connection on a windows laptop running the Brainvision recorder software (Brain Products GmbH, Gilching, GER). After the experiment the data from the SD card were transferred to the laptop's hard drive with the help of the LiveAmp File Converter software (Brain Products GmbH, Gilching, GER). FCz was used as online reference and AFz served as the ground. Head movements were recorded simultaneously with the help of the LiveAmp's built-in gyro sensors.
Data processing. The recorded and converted EEG data were pre-processed offline using custom Matlab and EEGLab scripts 67 . Separate pre-processing parameters and routines were used for event-related potentials (ERP) and event-related spectral perturbations (ERSPs). In general, the pre-processing was handled similarly to lab-based experiments. Previous research showed that artefactual data is produced mainly because of mechanical cable displacement and cable crossing 64 , which we prevented by proper cable management and cable relief mechanisms. Also, strong displacement artefacts were formerly shown to be elicited only by exceptional movement intensity like in a running task 14 . Comparable data quality and P300 classification for indoor seated and outdoor movement conditions were reported by Debener et al. 13 as early as 2012. Additionally, ERPs and ERSPs should not be prone to direct movement artefacts due to the averaging process. Only when a movement is locked to a specific point in time related to the task this should pose a problem to the validity of the outcome.
For ERPs, the data were bandpass-filtered between 0.5 and 20 Hz using a fourth-order IIR Butterworth filter with DC-offset removal. For ERSP analysis, data were bandpass-filtered between 0.5 and 30 Hz. Then, two statistical channel rejections were performed with the first excluding all channels with a kurtosis of 8 SD or higher and the second excluding all channels with a probability higher than 5 SDs. These channels were excluded from both the ERP and ERSP dataset. Then the data were average-referenced to the remaining electrodes. On average, 2,2 channels were excluded per subject.
In a next step, the ERP data were sampled down to 125 Hz and saved in a separate dataset solely for the purpose of independent component analysis (ICA). The downsampling was necessary to facilitate ICA computation. Before running the IC decomposition, data from the ERP and ICA dataset were segmented into epochs ranging from −0.3 to 1.2 seconds relative to the stimulus onset.
For the ERSP data, the signal was sampled down to 250 Hz and epochs were created ranging from −1s to 2 s regarding stimulus onset. The downsampling was executed due to data handling purposes and the facilitation of further computations.
With the help of an automated trial rejection (voltage threshold: 1000 µV, probability threshold: 5 SD, maximum percent of total trials to reject per iteration: 10%), artefactual trials were detected in the ERP dataset and subsequently deleted from all separate datasets (ERP, ICA, ERSP). After trial rejection, an extended ICA was performed on the ICA dataset.
Then, IC-weights were transferred back to the original ERP-and ERSP-data. To reduce individual bias in data cleaning, independent components (ICs) representing artefacts were detected via the ADJUST plugin 68 . These artefactual ICs were then deleted from the ERP and ERSP data. Subsequently, previously excluded EEG-channels were spherically interpolated.
Questionnaires and behavioural data. For subjective workload evaluation a German translation of the NASA-TLX was used. Raw scores without factor weighting were passed on for statistical evaluation for each of the six dimensions. Behavioural responses were calculated as the latency difference between the stimulus-onset and right-hand button press. Only correctly detected target trials were considered for subsequent analyses. A trial (2019) 9:13086 | https://doi.org/10.1038/s41598-019-49503-4 www.nature.com/scientificreports www.nature.com/scientificreports/ was counted as correct, if the participant pressed the button following a target stimulus. Further analyses regarding incorrect responses and omission errors were not pursued, given the small proportion of errors (max 2.7%).
Walking data was analysed using the amplifier's gyro sensor data. For walking and the obstacle course conditions, after band-pass filtering between 0.1 Hz and 20 Hz, average stepping frequency was detected by identifying the power-peak of the gyro acceleration between 0.5 Hz and 3 Hz. Then the original gyro data was filtered using a wavelet at this peak-frequency. At last, positive peaks of this filtered signal were used as step-markers for calculating number of steps and step interval variability.
EEG signal processing and analysis. ERPs: Using the EEGLab study design, correct standard and target trials were averaged for the six conditions across subjects. The baseline interval ranged from −200 ms to 0 ms pre-stimulus. For statistical analysis frontal N2-as well as parietal P2-, N2-, and P3-voltages were quantified as averages in specific time windows. To obtain specific frontal N2 and parietal P2 and N2 time-windows, grand-averages were calculated for each movement complexity condition's target trials irrespective of target probability. By detecting the maximum voltage for these grand average ERPs frontal N2 between 150 ms and 300 ms relative to stimulus onset, the N2 time window was determined by placing a 50 ms window around the peaks (stand: 264 ms +/−25 ms, walk: 266 ms +/−25, obstacle course: 266 ms +/−25 ms). The N2 was then parameterized as the mean amplitude in these time windows for each factor combination. The same procedure was performed for parietal P2 (search between 100 ms and 200 ms) and N2 (search between 150 ms and 300 ms) to obtain grand average peaks. Distinct peaks were found for both the parietal P2 (stand: 126 ms, walk: 124 ms, obstacle course: 126 ms) and N2 (stand: 168 ms, walk: 176 ms, obstacle course: 178 ms). A time window of +/−20 ms was applied around the peaks to parametrize parietal P2 and N2 mean voltages. Due to the lack of temporal specificity, the P3 amplitude measure was parameterized as the mean amplitude in the time window ranging from 250 ms to 500 ms relative to the stimulus onset.
ERSPs: Similarly to ERP measures, only correct standard and target trials were used for ERSP computation. ERSP data was calculated using an EEGLab study. 28 Wavelets with three cycles and a cycle shortening factor of 0.5 were computed for frequencies between 3 Hz and 30 Hz. The baseline was set for the interval between −500 ms and −200 ms relative to stimulus onset. Theta power was averaged between 4 and 7 Hz before determining the movement complexity conditions' peak latencies analogously to the N2. The determined time-windows were 268 ms +/−150 ms for the standing condition, 256 ms +/−150 ms for the walking condition and 256 ms +/−150 ms for the obstacle course condition. Theta power was then averaged in these time windows for each factor combination.

Statistical analysis.
Statistical analyses were carried out using R version 3.4.3 69 and the ez-package 70 . For subjective and behavioural data 2 × 3 repeated measures analyses of variance (ANOVAs) were calculated using movement complexity condition and target probability as within factors. Regarding physiological measures, we used 2 × 3 × 2 repeated measures ANOVAs with movement complexity condition, target probability, and stimulus type (target/non-target) as within factors. For the testing of statistical significance, p-values equal to or below 0.05 were considered as significant. To account for family-wise error accumulation, significances within each ANOVA were corrected for false discovery rate (FDR) as indicated by Cramer and colleagues 71 . Post-hoc tests for movement complexity conditions were calculated using pairwise paired t-test comparisons for all factorial levels. Post-hoc test probabilities were also corrected for FDR. In cases of FDR-correction, adjusted critical p-values (p crit ) are provided. The ANOVAS' effect sizes are reported in η p 2 . Statistical plots and figures with a significant main effect of target probability display all possible factors; for measures without a significant TP main effect, plots were produced without splitting the graphs into TP-levels to facilitate comprehensibility and to declutter the plots.

Data Availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.