Cognition Impairment Prior to Errors of Working Memory Based on Event-Related Potential

Cognitive impairment contributes to errors in different tasks. Poor attention and poor cognitive control are the two neural mechanisms for performance errors. A few studies have been conducted on the error mechanism of working memory. It is unclear whether the changes in memory updating, attention, and cognitive control can cause errors and, if so, whether they can be probed at the same time in one single task. Therefore, this study analyzed event-related potentials in a two-back working memory task. A total of 40 male participants finished the task. The differences between the error and the correct trials in amplitudes and latencies of N1, P2, N2, and P3 were analyzed. The P2 and P3 amplitudes decreased significantly in the error trials, while the N2 amplitude increased. The results showed that impaired attention, poor memory updating, and impaired cognitive control were consistently associated with the error in working memory. Furthermore, the results suggested that monitoring the neurophysiological characteristics associated with attention and cognitive control was important for studying the error mechanism and error prediction. The results also suggested that the P3 and N2 amplitudes could be used as indexes for error foreshadowing.


INTRODUCTION
Performance errors may have serious consequences. For instance, the error made by Apollo astronauts almost led to the collision of the moon with their spacecraft (Shayler, 2000). Further, a serious mistake made by a driver may lead to casualties and loss of the means of transport. Therefore, it is important to study the errors for improving human-system safety. Many causes account for errors, with some defaults in cognitive function playing an important role. Investigating the error-related neural patterns is important to uncover the error mechanisms, thus helping in improving the safety.
A few studies explored the error-related neural mechanism (Ora et al., 2015;Shou et al., 2015) by analyzing multiple cognitive processes in one single task. Ora et al. (2015) analyzed the spatiotemporal neural activities during performance errors and found that attentional lapses and inappropriate action impulses caused subsequent performance errors in a d2 task. Shou et al. (2015) studied the error mechanism by the comprehensive analysis of N2 and alpha power in the Stroop task. The results revealed that the alpha power of error increased in the parietooccipital area under congruent trials, while the N2 amplitude decreased only in incongruent trials. These findings implied that poor sustained attention and poor cognitive control caused errors. As to working memory, what cognitive changes may cause errors is still unknown.
Memory updating is the key process of working memory. The P3 amplitude is associated with memory updating, and its latency is related to match function (Watter et al., 2001;Chen et al., 2008). On the contrary, P3 is closely linked with attention and cognitive resource reallocation. P3a is bound with attention to the selection-driven stimulus, especially the novelty stimulus (Falkenstein et al., 1994;Polich, 2007). In contrast, the P3b (Polich and Heine, 1996;Schapkin and Freude, 2013) is related to the reallocation of cognitive resources and memory. Some studies showed that a few cognitive resources were reallocated to other memory processes, resulting in a longer P3 latency and a decreased P3 amplitude (Polich, 2007;Daffner et al., 2011;Wild-Wall et al., 2011;Saliasi et al., 2013;Gajewski and Falkenstein, 2014) in working memory. One study also showed that latency jitter can decreased P3 amplitude (Aricò et al., 2014). P2 is part of memory information process related to the onset of memory updating (Lenartowicz et al., 2010;Yuan et al., 2016). N2 is regarded as an ERP component related to cognitive control in memory (Daffner et al., 2011;Gajewski and Falkenstein, 2014). N1 is the early component of information process, which remains sensitive to physical features of stimulus and reflects the recognition and code processes (Ritter et al., 1979;Hopf et al., 2002). To date, no studies have reported on the error mechanism of working memory. It is unknown whether memory updating, attention, and cognitive control may change in working memory and cause errors. No studies have identified error-related neural patterns in one single task for working memory. Therefore, the two-back working memory task was adopted in this study to investigate error-related neural patterns and cognition impairment.
A few studies showed the N1 difference in errors. Maybe, the stimulus is identical in correct and error trials. Also, it is too simple for recognition and hardly leads to error. Therefore, N1 was analyzed to verify the hypothesis in the present study. The memory updating is the key cognitive process in working memory. Therefore, the P3 and P2 amplitudes may decrease in error trials. Some studies (Ridderinkhof et al., 2003;Kok et al., 2004;Padilla et al., 2006;Schmajuk et al., 2006;Hanslmayr et al., 2007;Klimesch et al., 2007;Eichele et al., 2008;Mathewson et al., 2009;Mazaheri et al., 2009;O'Connell et al., 2009;Jensen and Mazaheri, 2010;Bode and Stahl, 2014;Ora et al., 2015;Shou et al., 2015) showed that poor attention and poor cognitive control were the error-related neural patterns, leading to changes in P3 and N2.
The P2 and P3 amplitudes may decrease in error trials, but the N2 amplitude may increase for impaired cognitive control due to the energy cost theory (Kok, 1990;Pinal et al., 2015).

Participants
This study was approved by the ethics committee of China Astronaut Research and Training Center. All participants provided written informed consent in accordance with the Declaration of Helsinki. For this study, 40 male participants, aged 19-34 years with an average age of 24 years [standard deviation (SD) = 2.92)] were enrolled. The participants were postgraduate or graduate students with normal or corrected-tonormal vision and without any reported psychiatric disorders. Five participants were excluded because of too many artifacts in electroencephalogram (EEG), and the remaining 31 participants who had more than 20 performance errors were selected for further analysis.

Two-Back Working Memory Task
A numerical two-back task was used in the present study (Yuan et al., 2016) The sequences of 1, 2, 3, 4, 5, 6, 7, 8, and 9 were pseudo-randomly displayed in a white against black background on a 24-inch LCD screen (with an updating rate of 60 Hz) one by one. The height and width of the stimulus were 1.8 and 1.4 cm, respectively. At the beginning of each trial, a stimulus (Figure 1) was displayed for 500 ms or disappeared if responded. Then, a new stimulus was displayed after 2500 ms. Participants were required to press the "F" or "J" button as accurately and quickly as possible. If the stimulus presented was the same as the one that had appeared two presentations before, the participants were required to press the button "F"; if not, the participants were required to press the button "J." The time of the task was 10 min, and the target rate was 33.33%. The number of trials was different between participants, probably due to the reaction times. However, the minimum number was 200 from the required time in the task design.

Procedure
The participants were needed to come to the laboratory two times. The first time was to get training, 1 day before the test, to ensure that all participants were familiar with the task. The second time was to get the formal test the next day. EEG data were recorded during the working memory test. The laboratory had a shield room. The participants were seated in the shield room to record the EEG data. The size of the shield room was 2.4 m × 2.4 m × 2.4 m.

EEG Recording
Electroencephalogram data were recorded during the working memory test. After placement of electrodes, the participants were requested to sit in a comfortable chair 80 cm away from the LCD screen and instructed to refrain from excessive blinking and movement during data collection. Participants then underwent a two-back working memory task. Then, the EEG was recorded with Ag-AgCl electrodes embedded in an elastic cap (EasyCap, Brain Products GmbH) at 63 scalp locations according to the 10-20 system of Jasper (1958). The vertical electrooculogram (vEOG) was recorded from an electrode infraorbital to the left eye and the horizontal EOG was recorded at the outer canthi of the right eye. The ground electrode was positioned at AFz, and the reference electrode was positioned at FCz. The Vertical and horizontal electrooculogram (EOG) were recorded from electrodes above and below the right eye and on the outer canthi of both eyes. The electrode impedance was kept below 5 k . Signals were recorded using the BrainVision Recorder (Brain Products GmbH, Ver. 1.03), with a band-pass filtered at 0.01-250 Hz. The sampling rate was 1000 Hz, and the signals were amplified in the range of ±3.27 mV, and at a resolution of 0.1 µV.

Statistical Analysis of Behavioral Data
Descriptive statistics of error rate, reaction times (RTs) of correct and error responses, and the total number of errors were computed. The error was that the response did not accord to the stimulus. For example, although the button accorded to the stimulus was "F", the participant pressed "J"; this was an error response. The omission was not included in error statistics. The error rate was defined as the number of error responses divided by the total number of responses. RTs of two trials (correct vs. error) were examined using the paired t-test (Hsieh et al., 2010;Murphy et al., 2010;Sambrook and Goslin, 2014;Ora et al., 2015;Shou et al., 2015).

EEG Analysis
The EEG data were processed off-line using the BrainVision Analyzer 2.0. Software (Brain Products GmbH, German). The mastoids (TP9, TP10) were selected as the new references (Brain Products GmbH; Möckel et al., 2015). The semi-automatic inspection method was implemented to inspect the raw data. The gradient criterion was set at 50 µV/ms. The maximum absolute difference allowed was 200 µV with a 200-ms interval. The amplitude was in the range of −200 to 200 µV. The allowed activity with the lowest amplitude was 0.5 µV. After inspecting the raw data, the EEG signals were corrected for eye movement artifacts using the artifact rejection method, which was based on the Gratton and Coles' algorithm. This procedure was implemented in the BrainVision Analyzer 2.0 Software. The continuous EEG signals were filtered using a band-pass filter from 0.1 to 35 Hz with a 0-phase shift of 48 dB, and the notch filter was 50 Hz.
Stimulus-locked data were segmented into epochs of −200 ms to 800 ms after stimulus and were baseline-corrected relative to the interval of −200 ms to 0 ms. The ERPs of accepted trials were then averaged separately for obtaining a correct or error trial. The N1, P2, N2, and P3 values were quantified as the maximum amplitudes, and the intervals were increased from 130 to 170 ms, 170 to 270 ms, 270 to 360 ms, and 400 to 500 ms post-stimulus, respectively. The maximum amplitudes of each ERP component were the average of 10 points around the peak amplitude. Grand average waveforms of N1, N2, and P3 were created for the correct and error trials at each of the three electrode sites (Fz, Cz, and Pz). The three components were analyzed using a two-way repeatedmeasures analysis of variance with Trial (correct and error) and Site (Fz, Cz, and Pz) as within-subject factors (Gehring and Fencsik, 2001). With no overt P2 in Pz, ANOVA with repeated measures was conducted on P2 with Trial (correct and error) and Site (Fz and Cz) as within-subject variables. The partial eta squared (η 2 p ) was the effect size, and the Greenhouse-Geisser correction was used where appropriate (Liu et al., 2015). Paired-sample t-tests were used for analysis of latencies of Trial (correct and error).

Behavioral Results
Due to different RTs, the response number of the participants was between 200 and 422 within 10 min and the mean response number was 356.4 (SD = 59.8). The mean error rate was 0.12 (SD = 0.05) with the 0.01 (SD = 0.02) omission rate, and the mean error responses in 31 participants was 43.8 (SD = 18.6). The RTs of error trials were significantly longer than those of correct trials (t = −2.078; P = 0.01). The results are shown in Figure 2.

ERP Results
The means of ERP amplitudes and latencies are shown in Tables (Figure 3) showed that the amplitudes in correct trial were far larger than the errors, especially in the central and parietal areas. These results were also illustrated using topographic maps (Figure 4).

DISCUSSION
The present study partially verified the hypotheses. Reduced P2 and P3 amplitudes in error trials were found. However, the N2 amplitudes increased in error trials. The RT of the error trial was longer than that of the correct trial. The results revealed that poor attention, poor cognitive control, and impaired memory   updating might cause errors in working memory. The N1 amplitudes showed no difference between the two trials. As expected, the N1 amplitudes showed no significant changes in different trials. N1 is thought to be the early component of information processing. Its amplitudes reflect the recognition and code processing of stimulus (Kutas and Hillyard, 1980;Hopf et al., 2002), which are sensitive to the physical features of the stimulus. Therefore, the results indicated that the neural activity was comparable during the stimulus encoding stage of working memory process between correct and error trials. Additionally, no differences were found between trials that resulted from the same too simple physical features of the stimulus.
No study has been found regarding the association of P2 with an error. The results revealed that the P2 amplitude decreased when participants presented an error response. Previous studies found that the P2 component reflected the initial stage of context updating, and was believed to be the onset of memory updating (Lenartowicz et al., 2010;Yuan et al., 2016). Therefore, the diminished amplitudes of error trials were related to the impaired ability of memory updating onset. The onset of updating was encoded after stimuli and then translated to phonological representations. Thus, it was considered one of the crucial steps for successful performance and impaired ability, resulting in an error response in this study.
The increased N2 amplitude in error trial implied that the cognitive control decreased (Kok, 1990;Caseras et al., 2006;Unsworth et al., 2009;Bode and Stahl, 2014;Pinal et al., 2015) and induced errors based on the energy cost theory (Kok, 1990;Pinal et al., 2015). N2 was related to cognitive control, including response selection and conflict detection in memory (Daffner et al., 2011;Gajewski and Falkenstein, 2014). The peak N2 amplitude was significantly larger in the older group than in the younger group (Pinal et al., 2015), which might be due to the use of more energy for the older participants. The increased N2 amplitude reflected an enhanced energetic cost in cognitive control, such as mismatch and response selection Folstein and Van, 2008;Daffner et al., 2011;O'connell et al., 2012b). Previous studies showed that a high N2 amplitude was found in patients than in controls because the patients need more resources to process the information including the mismatch and response selection (Bruder et al., 1998(Bruder et al., , 2001Daurignac et al., 2006;Guillem et al., 2006;Shu et al., 2014;Sumich et al., 2014;Pinal et al., 2015;Zuj et al., 2017). A larger N2 amplitude possibly demanded greater efforts in task completion and even induced hyperactivation by way of more efforts during the completion of a relatively simple task (Bruder et al., 1998(Bruder et al., , 2001Guillem et al., 2006;Sumich et al., 2014). Some studies showed a diminished P3 amplitude in working memory because more resources were recruited in the cognitive control and increased the N2 amplitude (Daffner et al., 2011;Gajewski and Falkenstein, 2014). More neuron-related cognitive control was activated, subsequently increasing the N2 amplitude. The stimulus was identical in all the error and correct trials, and therefore the cognitive demands remained the same. This indicated that more resources were used for the same task, and the error rate increased, indicating a declination in cognitive ability in the wrong trial.

Decreased P3 Amplitude Reflected Impaired Attention
Falkenstein and Polich (Falkenstein et al., 1994;Polich, 2007) showed that frontal P3(P3a) was related to attention selection (Polich and Heine, 1996;Schapkin and Freude, 2013) and parietal P3(P3b) was associated with cognitive resource reallocation. The decreased P3 amplitude in frontal and parietal areas implied that some resources were reallocated in other memory processes, impairing the attention. Gajewski and Falkenstein (2014) found that cognitive resource was reallocated in other processes and the impaired attention led to a decreased P3 amplitude in the two-back working memory task (Polich and Heine, 1996;Polich, 2007;Daffner et al., 2011;Wild-Wall et al., 2011;Saliasi et al., 2013;Schapkin and Freude, 2013). A decline of P3 amplitude in frontal and parietal areas implied impaired attention and diminished cognitive resources, resulting in increased omission rate, false alarm, and more errors (Donchin and Coles, 1988;Polich, 2007;Zhao et al., 2013;Schapkin and Freude, 2014). On the contrary, longer RT in error trials was thought as attention lapses, which was supported by fMRI (Weissman et al., 2006) and other findings (Padilla et al., 2006). This was consistent with the results of the present study, which implied impaired attention.

Decreased P3 Amplitude Reflected Deficit Ability in Memory Updating
The P3 amplitude is an effective measurement of memory updating (Donchin and Coles, 1988;Watter et al., 2001;Polich, 2007;Chen et al., 2008;Zhao et al., 2013), and the latency is related to the match function (Donchin and Coles, 1988;Watter et al., 2001;Polich, 2007;Chen et al., 2008;Zhao et al., 2013). The ability associated with memory updating increased with better performance, leading to an increased P3 amplitude in the parietal area when the participants underwent n-back training (Watter et al., 2001;Chen et al., 2008). A reduced amplitude indicated that the memory updating capability was impaired. Updating included code, operation, search, and selection of information. Impaired memory updating caused some faults in information operation and updating in the memory mode. Thus, the errors were related with poor memory updating.

CONCLUSION
Diminished P3 and P2 amplitudes and an increased N2 amplitude in errors were related to impaired attention and deficit in memory updating and cognitive control. The diminished P3 amplitude indicated that more cognitive resources were reallocated to other memory processes, leading to impaired attention in the memory updating process. As a result, it caused a deficit in the ability of memory updating and decreased cognitive control. This was the first study on the poor memory updating causing errors.
Poor attention and cognitive control were the major causes of error in many common tasks (Ridderinkhof et al., 2003;Kok et al., 2004;Padilla et al., 2006;Schmajuk et al., 2006;Hanslmayr et al., 2007;Klimesch et al., 2007;Mathewson et al., 2009;O'Connell et al., 2009;Jensen and Mazaheri, 2010). Hence, much attention should be paid to the neurocognitive characteristics, which are important for error mechanism and can be used as indexes for error prediction. How to predict the error based on EEG trial-totrial analysis needs further investigation in the future.