The Effects of Amnestic Mild Cognitive Impairment on Go/NoGo Semantic Categorization Task Performance and Event-Related Potentials

INTRODUCTION

It is well recognized that individuals with a diagnosis of amnestic mild cognitive impairment (aMCI) are at higher risk of developing Alzheimer’s disease (AD), one of the most common forms of dementia [1, 2]. While impairment in episodic memory is a hallmark symptom of aMCI, there is accumulating evidence that individuals with aMCI also have impairments in other cognitive domains, including executive function [3–9].

Executive function, also referred to as cognitive control, encompasses top-down cognitive processes that guide behavior. These cognitive processes include stimulus evaluation, response execution and inhibition, conflict monitoring, planning, organizing, and decision making [10–13], all of which play a role in day-to-day activities. To characterize cognitive control and its deficits in individuals with MCI, a variety of clinical (e.g., Behavior Rating Inventory of Executive Function, the Stroop color-word task, Wisconsin Card Sorting Test, Trails B, Letter fluency) and experimental measures (e.g., the Brown-Peterson task, the Hayling task, Erickson flanker task, the Simon task, Go/NoGo task, the stop signal task) have been used [14–24]. Here we report on electrophysiological correlates of cognitive control involved in execution and inhibition of a response in a cohort with aMCI and a group of cognitively normal controls.

Response inhibition entails overriding an ongoing course of action in favor of an alternative course [25]. As we navigate our ever-changing environment, we evaluate stimuli, monitor conflict between current actions or intentions in the face of upcoming and competing sources of information, and appropriately execute or withhold a response [26, 27]. For instance, when walking down a street, we stop if we see an approaching car but continue to walk if we simply see other people walking by. There is general consensus that individuals in early stages of AD have impaired response inhibition (see [28] for review; [29–36]). What remains unclear is whether response inhibition deficits occur in the pre-dementia stage, as existing studies have yielded inconsistent results [14]. While some behavioral studies in MCI patients have found response inhibition deficits on Erickson flanker [23], Stroop [15, 22], and stop signal tasks [24], others have observed no differences between normal controls and individuals with aMCI on negative priming, Stroop, Go/NoGo tasks [37], and the classic Hayling task [16]. Differences in tasks and stimuli across studies, in addition to the heterogeneity in the diagnosis of MCI, may explain discrepancies in the behavioral findings. Given that neural alterations in aMCI and mild AD precede and eventually lead to behavioral deficits [38–42], we sought to evaluate neural responses underlying behavioral changes in response inhibition using electroencephalography (EEG) to better characterize changes in cognitive control in individuals with aMCI [43, 44].

Many EEG-based studies of normally aging individuals have used the Go/NoGo paradigm to study response inhibition [45–50]. The standard Go/NoGo task requires subjects to respond to a certain type of stimuli (Go) and inhibit/refrain from responding to another type of stimuli (NoGo), pre-defined by a specific set of rules. Typically, the Go/NoGo paradigm yields two event-related potential (ERP) components: an early fronto-central negative deflection developing between 200 and 400 ms post-stimulus onset (N2) and a later positive component (P3) between 300 and 600 ms [51–55], which are differentially modulated by the Go and NoGo trials. The exact functional significance of each of these two ERP components is debated. The suggested roles of N2 vary from being a marker of response inhibition ([56–58]; also see [46, 59] for reviews) to stimulus evaluation and conflict monitoring [26, 61]. On the other hand, the suggested roles of P3 vary from being an index of evaluation of stimuli and attentional allocation [62, 63] to response monitoring (e.g., [52]). Nevertheless, there is general consensus that both N2 and P3 are modulated during response execution and inhibition [64, 65].

To the best of our knowledge, only one EEG study has thus far examined response execution and inhibition in aMCI using the Go/NoGo paradigm [19]. Cid-Fernández et al. [19] compared ERPs of 63 healthy controls and 30 aMCI participants obtained during a Go/NoGo task. They found that aMCI patients had smaller Go-N2 and NoGo-N2 amplitudes, longer response times (on Go trials), and more errors compared to controls, suggesting an overall impairment in performing the Go/NoGo task. This Go/NoGo study and other ERP studies that have used the oddball paradigm in aMCI [66–68] have predominantly examined perceptual categorization using stimuli such as tones of different frequencies, numbers, letters, and shapes. Although response inhibition in the real world is often based on how objects we encounter are categorized (e.g., when walking through the woods, deciding whether to take a step or not when a rope versus a snake is encountered), to our knowledge, none have studied the Go/NoGo paradigm in aMCI in the context of semantic categorization. Given that semantic memory deficits have been observed in behavioral [69–74] and ERP [75, 76] studies in aMCI, examining response execution and inhibition in the context of semantic categorization could provide useful information as to whether these operations interact in individuals with aMCI.

Semantic categorization (e.g., categorizing items as animate versus inanimate) allows for meaningful organization of objects in our surrounding environment [77–80]. In general, objects are categorized at three levels: basic, superordinate, and subordinate [81]. For example, a ‘dog’ at the basic level can be categorized at the superordinate level as an ‘animal’ and at the subordinate level as a ‘golden retriever.’ Basic and superordinate categorizations are more frequently involved in day-to-day functioning than subordinate categorization, which involves more in-depth/expert knowledge (e.g., knowledge used by dog experts, car experts, bird experts). Previously, we examined the effects of age on cognitive control using Go/NoGo tasks involving basic and superordinate categorization in an EEG study by comparing cognitively normal young and older adults [47]. We found that cognitively normal older adults had reduced NoGo-N2 and NoGo-P3 amplitudes compared to younger adults, suggesting alterations in neural processing during response inhibition. Additionally, we found that older adults had longer N2 latency for superordinate compared to basic categorization; in contrast, younger adults showed no N2 latency differences between the two categorization tasks, suggesting that complexity of categorization modulates neural processing in older adults in early stages of processing. However, both groups exhibited similar task effects in later stages of neural processing with shorter P3 latency and larger P3 amplitude for basic compared to superordinate categorization. The logical next step was to examine how age-related cognitive control mechanisms differ from disease-related mechanisms such as in individuals withaMCI.

The goal of the current study was to determine whether and how aMCI affects the ERP correlates (N2/P3) of cognitive control (response inhibition and/or response execution) during semantic categorization, using the same paradigm from our earlier studies [47, 83]. The objective ERP markers isolated in this study may help detect early neural alterations corresponding to changes in cognitive control in individuals with aMCI, and in the future, could be applied as a measure to monitor disease progression and treatment response given the high predictive value of cognitive control deficits in everyday functioning [7, 84–86]. Based on our previous work using the Go/NoGo paradigm and in consideration of the work by others cited above, we predicted delayed latency and/or reduced amplitude in N2 and/or P3 components in aMCI participants compared to controls. We also expected aMCI-related ERP changes (delayed latency and/or reduced amplitude) to be more prominent during superordinate compared to basic categorization and during NoGo compared to Go trials.

METHODS

RESULTS

DISCUSSION

We compared a group of individuals diagnosed with aMCI to a group of cognitively normal controls in order to examine the effects of aMCI on behavioral measures and neural markers related to response execution and inhibition (the N2 and the P3 ERP components) across two Go/NoGo tasks that varied in levels of semantic categorization (basic and superordinate). Behavioral data revealed that individuals with aMCI made more commission errors on the NoGo trials compared to controls, whereas both groups had comparable omission errors during the Go trials. Individuals with aMCI had significantly prolonged N2 ERP latency compared to controls in both tasks across conditions. Both groups had prolonged N2 and P3 latency and reduced P3 amplitude during superordinate compared to basic categorization, and both had prolonged N2 and P3 latency and increased P3 amplitude during the NoGo trials compared to the Go trials.

More commission errors or false alarms on the NoGo trials in individuals with aMCI compared to the controls suggests a potential deficit in response inhibition ([56]; also see [59]). However, since we used unequal trial distribution (80% Go and 20% NoGo) to enhance pre-potent response tendencies, difficulty in identifying the less frequent NoGo trials or degraded ability to resolve response conflict during the less frequent NoGo trials [26, 99] may have also contributed to these findings. The study by Cid-Fernández et al. [19] that examined response inhibition using a Go/NoGo task found an overall reduction in response accuracy in the aMCI group compared to the control group. However, they did not examine commission and omission errors separately, and thus, our findings cannot be directly compared to their study. The reduced NoGo accuracy findings in the aMCI group were corroborated by the deficits observed on the neuropsychological measures of cognitive control. Overall, the aMCI group performed significantly less well compared to the control group not only on neuropsychological measures of episodic memory (logical memory immediate and delayed) but also on measures of cognitive control (Trails B and letter fluency), as has been observed by others [18, 66]. A study by Traykov et al. [22] found that aMCI participants had increased errors during the Stroop interference condition compared to controls, supporting deficits in cognitive control similar to that observed in our NoGo accuracy data.

We found prolonged N2 latency in the aMCI group compared to the control group. A recent meta-analysis by Howe [100] of ERP studies in MCI and AD patients revealed that N2 latency is a consistent marker of cognitive deterioration, in line with our findings. For instance, Bennys et al. [101] found that frontal N2 latency discriminated between MCI and controls in 80% of the cases. Furthermore, studies have also revealed that N2 latency is an accurate and sensitive measure of conversion from MCI to AD [67, 102]. Papaliagkas et al. [67] found that an N2 latency cut-off of 287 ms for auditory evoked potentials obtained using an oddball paradigm predicted conversion from MCI to AD with 100% sensitivity and 91% specificity in their sample. However, our N2 latency findings diverge from that of the Go/NoGo ERP study by Cid-Fernández et al. [19], who observed N2 amplitude but not N2 latency differences between the aMCI and control groups. Several reasons might have led to the difference in findings. The visual Go/NoGo stimuli in the Cid-Fernández et al. [19] study were preceded by auditory stimuli. Participants were required to ignore the distracting auditory stimuli as they actively responded to the visual Go/NoGo stimuli, which would have engaged cognitive processes other than those involved in performing a Go/NoGo task without distraction such as ours. Furthermore, their study examined perceptual categorization using letters, numbers, and shapes, whereas we examined categorization of objects.

Although we observed an interaction between response type and group for error rates (significant group difference in commission error rates but comparable omission error rates in both groups), we did not observe such an interaction between response type and group for N2 latency. The overall N2 latency was delayed in both Go and NoGo conditions in the aMCI group compared to the controls. While delayed NoGo-N2 latency may explain increased commission errors (false alarms) in the aMCI group compared to the controls, delayed Go-N2 latency in the absence of increased omission errors or prolonged RT on Go trials in the aMCI group is puzzling. It is possible that the neural mechanisms that support cognitive operations common to both Go and NoGo trials are affected in aMCI. Different cognitive operations, including inhibitory processing and conflict monitoring [26, 61], have been suggested in relation to N2 associated with the Go/NoGo task [26, 61]. In particular, an independent component analysis of N2 by Kropotov et al. [58] has revealed three independent cognitive operations. These include sensory preparatory set, conflict monitoring, and response inhibition. While our Go/NoGo tasks were not designed to parse out the exact functional correlate that corresponds to N2 in relation to these three cognitive operations, a common cognitive operation involving Go and NoGo trials in our task was the sensory preparatory set or initial evaluation of the stimuli based on the pre-set rules. It is plausible that aMCI slows neural processing related to the initial evaluation of the stimuli resulting in prolonged overall N2 latency during both Go and NoGo trials compared to controls. Prolonged Go-N2 latency in the absence of corresponding Go accuracy deficits may be an early sign of neural degradation. This decline in function may affect response execution at the behavioral level as the disease progresses or may affect performance when conflict monitoring during response execution is more demanding (e.g., using more complex response-stimulus criteria). These speculations are based on the suggestion that neural alterations in aMCI and mild AD precede and eventually lead to behavioral deficits [38–40, 42]. Also, prolonged NoGo-N2 latency and increased false alarm rates suggest that additional neural mechanisms that support response inhibition may also be affected by aMCI. Although this mechanism is indexed by N2, it might involve a different neural generator [58]. We cannot determine these from our data, and future studies involving paradigms that can parse out the functional cognitive operations related to the Go/NoGo task and their source are necessary to evaluate these possibilities.

Both groups exhibited similar task effects for the reaction time measure. Behavioral response in the basic categorization task (SiC) was faster compared to the superordinate categorization task (ObA) in both groups, supporting the findings of previously conducted behavioral studies on categorization of typical objects [103–107] with the exception of those that have examined ultra-rapid object categorization (e.g., [108–111]). It has been suggested that basic categorization is faster since it depends on global, coarse perceptual discrimination [112] compared to superordinate categorization, which depends on semantic information as well as perceptual information, as long as sufficient time is allowed to process information.

The ERP findings related to task effects including prolonged N2 and P3 latencies and reduced P3 amplitude in the superordinate compared to basic categorization task suggest that both normal older adults and individuals with MCI respond similarly to task modulations. While younger adults exhibit task modulations in P3 latency and P3 amplitude similar to normal older adults and individuals with MCI, they do no exhibit task related N2 latency differences [47, 82]. These findings suggest that N2 latency provides unique insights into the differential effects of age on neural processing related to categorization, as observed previously [47].

Our prediction of greater impairment in superordinate compared to basic categorization in MCI was not supported by the findings. Although we observed impairment in letter fluency on neuropsychological assessment, we did not observe any deficits on category fluency in our cohort. It is possible that the aMCI individuals included in our study did not have semantic memory impairment that was severe enough to interfere with performance on the superordinate task more so than the basic task. The impact of categorization level may become more obvious with disease progression or if response time limitations are placed.

Also, in the current study, higher N2 amplitude was observed during superordinate compared to basic categorization in both control and aMCI groups, whereas in our previous study, the N2 amplitude did not significantly differ across tasks although a comparable trend was noted. Larger N2 amplitude in the superordinate categorization condition could be due to a strategy preference toward meaning-based processing of information [113] in the early stages of processing in older adults. However, studies that have examined semantic categorization in young adults have observed similar N2 amplitude differences between superordinate and basic categorizations, suggesting that this is not just a strategy preference. A study by Tanaka et al. [114] that required participants to make true/false decisions by matching between an individual category name (basic, superordinate, and subordinate) and a subsequently presented image found larger N2 amplitudes for superordinate compared to basic categorization in the frontal sites. They suggested that the frontal N2 amplitude difference reflects differential semantic processing between superordinate and basic levels. It is possible that N2 amplitude task effects were not observed in our previous study due to relatively smaller subject numbers in each group compared to our current study. Shorter P3 latency and larger P3 amplitude were observed in SiC compared to ObA in both groups irrespective of the response types, consistent with the findings of the Maguire et al. [82] and Mudar et al. [47] studies, suggesting that irrespective of age effects or early disease effects in MCI, the categorization levels modulate P3 similarly, and the same follows for N2.

As far as the effects of response type, both groups exhibited prolonged N2 and P3 latency and increased P3 amplitude during NoGo trials compared to Go trials. While these findings are consistent with what is typically observed in a Go/NoGo paradigm [45, 115], we had anticipated that the aMCI group would show more pronounced changes during NoGo trials compared to Go trials, consistent with our behavioral data. N2 latency differentiated the aMCI and control groups irrespective of the response type. It is likely that accuracy differences between Go and NoGo trials may be better captured by other EEG measures such as event-related spectral perturbations [116]. ERPs are time-locked neural responses obtained by averaging across multiple trials of interest. The neural activity captured by ERPs constitutes only a limited dimension of the total neural response to a stimulus and does not include the non-phase-locked activity at different frequencies [117–120]. Perhaps underlying differences in spectral perturbations at a given frequency band (e.g., alpha) may characterize underlying accuracy differences observed between Go and NoGo trials in the aMCI group. Subsequent studies are needed to explore if other EEG measure can extend our ERP findings. In general, our results related to response type suggest that neural processing is slower (indexed by prolonged NoGo-N2 and NoGo-P3 latency) and more effortful during response inhibition trials (enhanced NoGo-P3 amplitude). Whether this is related to more careful stimulus evaluation, conflict monitoring or inhibitory processing indexed by N2, or more rigorous response evaluation indexed by P3 during NoGo trials awaits clarification in future studies. Consistent with the findings of Mudar et al. [47], no differences were observed between Go-N2 and NoGo-N2 amplitude. Similar to normally aging adult controls, individuals with aMCI appeared to process both Go and NoGo trials with comparable effort, resulting in a lack of difference between Go-N2 and NoGo-N2 amplitudes. Although studies have observed differences in distribution of N2 and P3 across Go and NoGo trials [57], we did not examine differences in electrode distribution per se. Studies that examine N2 and P3 topographical distribution and source for Go and NoGo trials across MCI and control groups are further warranted.

In conclusion, individuals with aMCI manifest behavioral deficits and altered neural processing associated with cognitive control as related to the Go/NoGo task during semantic categorization. In particular, increases in commission errors and prolonged N2 latency differentiated the aMCI group from the controls. While this study provides an initial characterization of behavioral and electrophysiological correlates of cognitive control in a cohort with aMCI, changes with disease progression and changes as a result of clinical interventions await future investigations, as cognitive control is closely tied to functional abilities.