Executive function is a set of cognitive processes including attentional control, working memory and planning, and so on, and is necessary for selecting and successfully monitoring behavior (Alvarez and Emory, 2006; Chan et al., 2008; Diamond, 2013). Executive function is impaired in individuals with dementia, including those suffering from Alzheimer’s Disease (AD) (Collette et al., 1999; Perry and Hodges, 1999). Monitoring is a part of the executive function, and is the ability to monitor one’s own actions and responses during task performance in order to detect and correct errors. Impairment of executive function in AD patients is involved in the failure of self-monitoring (Ott et al., 1996), which may underlie subtle difficulties in coping with daily activities (Perry and Hodges, 1999). It has been suggested that deficits in self-monitoring occur after damage to the frontal lobes and other cerebral areas (Ott et al., 1996).

Feedback-Related Negativity (FRN) is a neurophysiological index that reflects the monitoring process associated with feedback inputs. FRN has been shown to be elicited by feedback stimuli (particularly negative stimuli) in a gambling task (Gehring, 2002; Hajcak et al., 2006; Holroyd et al., 2006) and a time production task (Miltner et al., 1997; Wild-Wall et al., 2009; Becker et al., 2014). This negative potential appears at a latency of 200–300 ms after feedback (Miltner et al., 1997; Falkenstein et al., 2000; Gehring, 2002; Holroyd and Coles, 2002; Nieuwenhuis et al., 2004), and is primarily distributed over the frontal-central scalp area. Generally, FRN means the negative peak obtained by subtracting the waveform for the positive feedback from that for the negative feedback, or the negative waveform obtained from the negative feedback. A great amount of evidence indicates that FRN is generated in the anterior cingulate cortex (Holroyd and Coles, 2002; Hajcak et al., 2007; Bellebaum and Daum, 2008; Holroyd et al., 2009). FRN codes negative prediction errors in the context of varying reword probabilities and magnitudes (Cohen et al., 2007; Bellebaum and Daum, 2008; Wu and Zhou, 2009; Bellebaum et al., 2010). As with other components of event-related evoked potentials (ERPs), many studies have revealed that FRN amplitude is reduced and its latency prolonged in the elderly compared to that in young individuals (Eppinger et al., 2008; Mathewson et al., 2008; Wild-Wall et al., 2009; Hämmerer et al., 2011; Pietschmann et al., 2011a), although some studies show no such age differences in FRN (Bellebaum et al., 2011). However, in line with the FRN aging effects, AD patients might show alterations of FRN. However, no reports have investigated the changes in FRN in AD patients.

Error-Related Negativity (ERN) is an ERP component elicited by an individual’s own behavioral errors. ERN measures may also provide insights to the monitoring system. Some reports have demonstrated that ERN amplitude decreases in older participants (Gehring and Knight, 2000; Falkenstein et al., 2001; Nieuwenhuis et al., 2002; Pietschmann et al., 2011a,b; Schreiber et al., 2011; Endrass et al., 2012), and is further reduced in AD patients with prolonged latency (Mathalon et al., 2003). Considering its similarity to ERN, this evidence suggests that FRN should also demonstrate decreased amplitude and delayed latency in AD patients.

P300 is a positive component at the dominantly parietal area approximately 300 ms after stimulus onset (Sutton et al., 1965), and is elicited typically by an oddball task, i.e., auditory, visual, olfactory or somatosensory stimuli (Yamaguchi and Knight, 1991a,b; Frodl et al., 2002; Polich, 2004; Bennys et al., 2007; Golob et al., 2007). This component reflects various cognitive processes, including context updating (Donchin, 1981; Polich, 2007), resource allocation (Wickens et al., 1983; Kramer et al., 1985; Polich, 2007), memory encoding (Karis et al., 1984; Fabiani et al., 1990; Johnson, 1995; Polich, 2007) and attention, stimulus evaluation, judgment and decision-making (Becker and Shapiro, 1980; Duncan-Johnson and Donchin, 1982; Kramer and Strayer, 1988; Katada et al., 2004; Gironell et al., 2005). There are many regions in the brain, especially in the parietal, temporal, prefrontal cortex and hippocampus, that contribute to its generation (Yamaguchi and Knight, 1991a; Halgren et al., 1998; Tarkka and Stokic, 1998; Kirino et al., 2000; Kiehl et al., 2001). P300 amplitude and latency are modulated by a variety of factors, including subjective probability of stimuli, stimulus saliency, availability of attentional resources (Kutas et al., 1977; Polich, 1986; Kramer and Strayer, 1988; Gonsalvez and Polich, 2002), and memory performance (Fabiani et al., 1990; Johnson, 1995). P300 has been proposed as one of the electrophysiological biomarkers of dementia (Olichney et al., 2011; Howe et al., 2014), and its usefulness has been well documented in the early diagnosis of dementia (Polich et al., 1990; Juckel et al., 2008; Ahiskali et al., 2009; Chapman et al., 2011; Vecchio and Määttä, 2011). Elderly people show a decreased amplitude and prolonged latency of P300, and these changes are more pronounced in dementia (Polich and Corey-Bloom, 2005; Bennys et al., 2007; Lai et al., 2010; Parra et al., 2012). We speculate that these findings regarding P300 may also be applicable to feedback stimuli.

Previous studies have also claimed that FRN and P300 encode different aspects of outcome evaluations (Yeung and Sanfey, 2004). These suggest that outcome evaluation can be roughly divided into two related processes: one is an early evaluation of the cognitive or motivational significance of feedback stimuli, which relates to FRN; and the other is a more elaborative evaluation of feedback stimuli, which is affected by the allocation of attentional resources such as intentionality or expectancy and is related to P300. Furthermore, other research has reported that FRN is sensitive not only to reward valence and magnitude, but also to expectancy towards reward magnitude. While P300 is sensitive to both feedback valence and reward magnitude, this sensitivity can be modulated by expectancy towards reward magnitude (Wu and Zhou, 2009). These findings suggest that FRN may play a role as a general mechanism that evaluates whether an outcome is consistent or inconsistent with expectations; whereas P300 is sensitive to the later, top-down controlled process of outcome evaluation, which links with the allocation of attentional resources, including reward valence, reward magnitude and magnitude expectancy. In AD patients, these processes of outcome evaluation for feedback stimulus are considered to be impaired.

Accordingly, we investigated the monitoring system of AD patients using a neurophysiological measure (i.e., FRN), in a gambling task. We hypothesized that, compared with healthy elderly people and healthy young people, the amplitude of FRN would be reduced, and latency delayed in AD patients due to disruption of feedback processing.

The aim of this study was to examine the changes in the monitoring system of AD patients and healthy control participants (HO) during gambling tasks using FRN. Results reveal that the AD group showed a larger amplitude and a delayed latency of FRN compared to the HO group, and the FRN amplitude correlated with cognitive function scores. This study revealed electrophysiological abnormalities of the monitoring function in AD. However, the increased amplitude of FRN in AD was unexpected. It is the opposite of predictions made from the original hypothesis.

The original hypothesis relating to changes in FRN for the AD group was based on prior evidence of aging effects on FRN. Many researchers have reported decreases in FRN amplitude associated with aging (Eppinger et al., 2008; Mathewson et al., 2008; Wild-Wall et al., 2009; Hämmerer et al., 2011; Pietschmann et al., 2011a), and our results also replicated this aging effect. ERN is also known to be affected by aging, with elderly people typically showing smaller ERN than young people (Gehring and Knight, 2000; Falkenstein et al., 2001; Nieuwenhuis et al., 2002). Several research groups have reported that the impairment of error processing and learning in older adults is the result of age-related changes in the mesencephalic dopamine system, such as the loss of dopamine receptors and the deterioration of dopaminergic receptor binding with aging (Volkow et al., 1998; Bäckman et al., 2000; Kaasinen et al., 2000). Based on findings that the efficiency of the dopamine system declines with age, Nieuwenhuis et al. (2002) extended reinforcement learning theory to older adults. They proposed that weakened phasic activity of the midbrain dopamine system leads to reduced negative reinforcement learning signals; implying that elderly people are learning impaired compared to younger adults (Nieuwenhuis et al., 2002). We had expected that this aging effect would be most pronounced in AD, which would mean that AD patients would exhibit smaller amplitudes in monitoring responses. A previous ERN study had indeed reported that amplitudes decreased for AD patients compared to healthy elderly people (Mathalon et al., 2003). However, in contrast to the results of that previous study, we found a significant increase of FRN amplitude in AD patients.

First of all though, whether AD patients in our study could understand the gambling task is a problem when we try to explain this result. Nevertheless, some behavioral studies have reported that AD patients have accomplished more complex tasks like the Iowa Gambling Task (Sinz et al., 2008). In our study, moreover, the switching ratio following the feedback was larger for the negative feedback than the positive feedback in all groups, suggesting that even AD patients could distinguish between positive and negative stimuli, and could avoid the options associated with recent negative results. Therefore, the negative component obtained from the differential waveform in AD could be considered as FRN.

As a result, we demonstrated that AD patients show a higher FRN. It is already known that the source of FRN is located in the ACC (Holroyd and Coles, 2002; Hajcak et al., 2007; Bellebaum and Daum, 2008; Holroyd et al., 2009), and abnormalities of the ACC in AD have been reported at histopathological and physiological levels. In AD patients, accumulation of both beta amyloid and tau (Morishima-Kawashima and Ihara, 2002; Leuba et al., 2009; Murphy and LeVine, 2010; Bloom, 2014), metabolic changes (Bracco et al., 2007; Lim et al., 2012), and decreased blood flows (Dukart et al., 2013; Long et al., 2013; Terada et al., 2013; Lin et al., 2014; Bailly et al., 2015) are observed in the ACC. The ACC is atrophied structurally (Buckner et al., 2005; Jones et al., 2006; Seeley et al., 2009; Krueger et al., 2010), and the anatomical connectivity with other areas is also impaired (Greicius et al., 2004; Rombouts et al., 2005; Wang et al., 2007; Boublay et al., 2016; Hafkemeijer et al., 2016b). On the other hand, higher functional connectivity of the ACC for the salience network has been reported in some resting-state fMRI studies for AD (Zhou et al., 2010; Hafkemeijer et al., 2016a). The salience network plays a crucial role in the detection of salient events from internal and external information (Seeley et al., 2007), therefore the network is inevitably involved in monitoring processing. The enhanced functional connectivity of the ACC in AD is dovetailed with the enhanced FRN in our study in the context of intensified detection of salient feedback information. Thus, the influence of AD pathology on the ACC function might be affected depending on the segregated functional network of the ACC.

One of the possible mechanisms for explaining this discrepancy is the compensation-related utilization of neural circuits hypothesis (CRUNCH; Grady, 2012). The idea of CRUNCH is that more neural resources are recruited by the elderly at low levels of cognitive load (when tasks are easier) than younger adults, who do not need them. At higher load levels, this compensatory mechanism is not effective, leading to less activation in elderly compared with young adults. This could apply to the relationship between AD and the healthy elderly (Grady, 2012). According to CRUNCH, enhanced FRN reflects a compensatory mechanism with larger recruitment of neural activity in AD patients compared to younger adults. Younger adults in our study might search a best strategy or hidden rule to get a better performance although the probabilities of positive and negative feedbacks were truly equal and the sequence was completely random.

Another idea that might explain the discrepancy is obtained from ERP studies of other diseases. A similar discrepancy between FRN and ERN has been reported in several psychiatric diseases in young people (van Meel et al., 2005, 2011; Groen et al., 2008; Holroyd et al., 2008; Vlamings et al., 2008; South et al., 2010; Larson et al., 2011; Santesso et al., 2011). Studies investigating FRN and ERN in children with autism have demonstrated that a robust FRN is equally elicited in children with autism and those with typical development (Larson and Clayson, 2011; Larson et al., 2011; McPartland et al., 2012; Stavropoulos and Carver, 2014), but ERN amplitude is significantly attenuated in autistic children (Vlamings et al., 2008; South et al., 2010; Santesso et al., 2011; Sokhadze et al., 2011, 2012; McMahon and Henderson, 2015). These results imply that individuals with autism may process external and concrete feedback normally, but have difficulty with the internal and more abstract regulation of performance. Similar neurophysiological mechanisms may be operative in AD patients.

As well as our FRN research in AD, there are several reports that have demonstrated FRN enlargement in other diseases. In depressed patients, augmentation of FRN amplitude has been reported (Tucker et al., 2003; Santesso et al., 2008; Cavanagh et al., 2011; Mueller et al., 2015). According to Beck’s cognitive theory of depression (Beck et al., 1979), depressed patients have a “negativity bias”, tending to focus more on negative information. This leads to abnormal responses to negative feedback, and yields an increase of the FRN amplitude. Although our study did not include AD patients with moderate or severe symptoms of depression, there was a significant relationship between FRN amplitude and depression scores only in AD. In AD patients, the “negativity bias” might be aggravated, even in mild depression.

Consistent with previous findings of P300 using an oddball task (Polich et al., 1990; Golob and Starr, 2000; Yamaguchi et al., 2000; Frodl et al., 2002; Bennys et al., 2007; Lee et al., 2013), the present study indicates that P300 exhibits decreased amplitudes and delayed latency in normal aging and AD. As expected, altered P300 was associated with cognitive function scores. The P300 responses are considered to be related to attention and memory processes (Polich, 2007); therefore, lower amplitudes and delayed latency of P300 in AD patients exhibits electrophysiological reflection of attention and memory deficits (Phillips et al., 2004; Bennys et al., 2007; Lai et al., 2010). It is considered that the causes of amplitude reduction in the AD group is atrophy of the hippocampus, decreases in blood flow, and decreases in functional binding. Because similar results were obtained in our gambling task, instead of an oddball task, the gambling task may be a more useful clinical tool in evaluating both FRN and P300 at the same time.

Our study has some limitations. First, we investigated only the alteration of FRN, but not of ERN. It has been speculated that generator sources of FRN and ERN components are located in the anterior cingulate cortex (Gehring et al., 1993; Dehaene et al., 1994; Gehring, 2002; Holroyd and Coles, 2002; Holroyd et al., 2002; Luu et al., 2003), and both potentials are associated with the monitoring system. FRN and ERN have been linked to the monitoring of externally provided and internally generated information, respectively (Müller et al., 2005). Therefore, in order to study abnormalities of the monitoring system in AD, it would be most interesting to examine the changes of both potentials simultaneously. Second, we have not explored any abnormalities of FRN in other cognitive impairment and dementia conditions, i.e., in mild cognitive impairment (MCI), frontotemporal dementia, or dementia with Lewy bodies. It is important to examine FRN changes in MCI, because MCI is a prodromal stage of AD. Several studies have reported that MCI patients show a reduction in P300 amplitude compared to healthy elderly people (Papaliagkas et al., 2008, 2011). Therefore, it is possible that similar ERP changes may occur in MCI, and thus whether FRN amplitude increases in MCI patients should be confirmed.

In sum, the present study demonstrated that the FRN in AD patients showed larger amplitude and delayed latency compared to age-matched controls, and correlated with depressive tendency. This indicates that enhanced monitoring response in AD patients might reflect a compensatory mechanism and/or negative bias in outcome evaluation. Psychophysiological measures in the feedback process could provide a clue to understand the neurobehavioral changes in AD patients.

EN, KO and SY designed the study and wrote the manuscript. EN, FI and RO collected the data. EN and KO performed the analysis. All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.