Neuroscience and Biobehavioral Reviews Understanding sources of adult age di ﬀ erences in task switching: Evidence from behavioral and ERP studies

The task-switching paradigm is a valid tool to measure age-related changes in executive functions. It allows identifying the most vulnerable cognitive control processes a ﬀ ected by aging. This review provides an overview about the current evidence on behavioral and electrophysiological signatures of adult age di ﬀ erences in task switching with a focus on age-related changes in ERP correlates of three task-processing phases: (1) advanced task preparation as re ﬂ ected by the cue-P3 and the CNV; (2) task implementation including P1 / N1, P2, N2, N450 / Ni as well as target-P3; and (3) response monitoring mechanisms as indicated by the Nc / CRN / MFN during correct responding and Ne / ERN in error trials. While most of these ERP correlates of executive control are reduced in older age, qualitative ERP di ﬀ erences between age groups are less consistent. We also report some recent ﬁ ndings from cognitive training research showing the potential for enhancement in task switching in older age. The results are discussed in the light of current models of cognitive control.


Brain alterations in aging
Aging is a life-long process beginning already in the uterus. The term aging has a negative connotation because of a substantial decline in various domains such as sensory, motor, and cognitive functioning. However, some of the age-related changes are positive and increase throughout the entire lifetime like crystallized intelligence, hitting its apex around age 60 or 70. In contrast, fluid intelligence enabling goal-directed behavior declines progressively beginning around age 30 or 40 (Horn and Cattell, 1967). Frontal and parietal brain areas associated with goal-directed behavior maturate late and begin to decline early (Braver and West, 2008;West, 1996). This decline with increasing age is accompanied by a decrease in volume of the brain tissue (Dennis and Cabeza, 2008;Raz et al., 2005). The prefrontal cortex associated with executive functions like flexible switching between tasks is particularly vulnerable to age-related changes (Braver and West, 2008). Besides neuroanatomical and morphological changes, age-related decline in cognitive performance is also due to impaired neuro-modulatory signaling and neurotransmitter availability, for instance, the reduction of dopamine release (Bäckman et al., 2000;2006;Braver and Barch, 2002). These changes seem to affect performance in older age, particularly during higher-order cognitive operations like switching among tasks.
The aim of the present review is to provide an overview over the existing literature on differences in task-switching behavior and its electrophysiological underpinnings between young and old populations as well as outlining some environmental and biological factors influencing task switching in older age.

Behavioral studies on age differences in task switching
The task-switching paradigm is particularly attractive because it enables to determine age-related changes in several control processes by applying only one experimental task. For instance, age-related changes in task preparation, interference and switching processes, as well as their interaction can be examined by using specific variants of the experimental paradigm (for a recent review, see Kray and Ferdinand, 2014). A majority of researchers have used an experimental task design that allows the computation of two types of costs related to task switching, termed mixing and switching costs in the following. In so-called mixed-task blocks, participants have to flexibly switch between two or more tasks. In addition, in so-called single-task blocks, performance is measured in a baseline condition in which the participants only have to perform one of the tasks (all-repeat). Mixing costs can be computed as the difference in performance between trials in https://doi.org/10. 1016/j.neubiorev.2018.05.029 Received 20 June 2017; Received in revised form 18 May 2018; Accepted 28 May 2018 mixed-task and single-task blocks (also termed global or general switch costs; cf., Kray and Lindenberger, 2000;Mayr, 2001), while switching costs are usually computed as the difference in performance between switch and repetition trials within mixed-task blocks (also termed local or specific switch costs). Mixing costs have been attributed to control processes required for maintaining multiple task sets in a switching situation, while switching costs are considered to reflect the reconfiguration of task sets in working memory (i.e., performing a task switch). Importantly, empirical evidence from structural equation approaches supports the view that both types of costs, mixing and switching costs, are correlated but can be separated and reflect distinct components of task switching (Kray and Lindenberger, 2000). Note that there are different preferred ways to compute mixing costs in the literature as well as theoretical explanations for their occurrence that are discussed in detail elsewhere (Grange and Houghton, 2014).
Long preparation for the upcoming task substantially reduces local switch costs but a portion of them cannot be fully eliminated even after additional prolongation of task preparation (Meiran, 2014;Rogers and Monsell, 1995). This so called residual switch cost was interpreted in terms of structural limitations of the cognitive system to switch between tasks in advance, task-set interference, or failure to engage in advance preparation (Allport et al., 1994;Rogers and Monsell, 1995). An important source of residual switch costs are so called "restart costs" in the first trial of a run after a short interruption by an instructional cue (Allport et al., 1994;Allport and Wylie, 2000;Monsell, 2017;Poljac et al., 2009). Restart costs are computed as the difference between the first and second or third target in the task sequence and are thought to result from some additional processing demands. Particularly, restart costs were attributed to cue-based task activation and retrieval, and are likely associated with task-set activation and resolving task interference (Altmann, 2007;Poljac et al., 2009; for further theoretical and methodological considerations see Kiesel et al., 2010).
1.2.1. The effects of advance preparation on age differences in task switching Successful performance during task switching requires mechanisms of proactive cognitive control including sustained activation and futureoriented use of predictive contextual information in a preparatory and selective manner (Kopp et al., 2014). The effects of advanced preparation have, for instance, been investigated by systematically varying the time to prepare for the next tasks (for a recent review, see Grange and Houghton, 2014). Results of most task-switching studies show reduced mixing as well as switching costs when subjects have more time to prepare for the next trial (Rogers and Monsell, 1995; for a review, see Meiran, 2014).
Most aging studies found no age differences in the effects of advance preparation. Younger and older adults usually show a similar reduction in mixing and switching costs when explicit task cues are provided (Cepeda et al., 2001;Hahn et al., 2004;Hartley et al., 1990;Kramer et al., 1999;Kray, 2006;Mayr and Liebscher, 2001;Meiran et al., 2001; but see Lawo et al., 2012), suggesting no specific deficit of older adults in processing and using task-cue information if they have enough time to prepare. Furthermore, Hahn et al. (2004) showed that both younger and older adults are able to use task cues as well as response cues for reducing switching costs. Also, under memory-based switching conditions, in which the task sequence has to be memorized and no explicit task cues are presented, both younger as well as older adults showed better task-switching performance with increased preparation time (De Jong, 2001;Gajewski et al., 2010a;Kramer et al., 1999;Kray, 2006;Kray and Lindenberger, 2000;Schapkin et al., 2014). In some studies, older adults even showed a larger reduction in switching costs than younger adults (Cepeda et al., 2001;Schapkin et al., 2014). However, it seems that older adults are sometimes not able to profit from prolonged preparation time under memory-based switching conditions, namely when the memory load is relatively high. For instance, in a study by Kramer et al. (1999), subjects had to keep track of four task repetitions before switching to the other task and here, older adults, but not younger adults, failed to profit from an increase in preparation time.

The influence of task interference on age differences in task switching
Task-switching performance is influenced by task interference induced by simultaneous maintenance of two or more task sets in working memory. Specifically, task interference means hampering of a full activation of one task by a second task that was recently used and is still active in working memory (Mayr and Keele, 2000).
Moreover, task-switching performance can be modulated by overlapping stimulus-response sets (congruency effect; Rogers and Monsell, 1995). On congruent trials, stimuli of both tasks are mapped onto the same response (e.g., the digit 3 is lower than 5 and odd and both are mapped onto the left response key). In contrast, on incongruent trials the same stimulus is associated with different responses in the two tasks (e.g., the digit 7 is greater than 5, requiring the right response key, and odd, requiring the left response key). A number of task-switching studies found switching costs to be greater on incongruent than on congruent trials, suggesting that switching from one task to the next is hampered when attributes of the other task have to be suppressed. Also, switching costs were larger when bivalent (associated with two tasks) rather than univalent stimuli (associated with only one task each) were used (for reviews, see Kiesel et al., 2010;Gade et al., 2014).
Interestingly, older adults seem to be insensitive to task interference when switching from one task to the next (e.g., Eppinger et al., 2007;Kramer et al., 1999;Kray et al., 2005; see also Lawo et al., 2012), but task interference seems to strongly influence age effects on mixing costs. For example, Mayr (2001) investigated whether age differences in mixing costs are dependent on task interference occurring at the stimulus and/or response level. Results of this study showed significant age differences in mixing costs under conditions of high stimulus ambiguity and response-set overlap. Age differences in mixing costs were reduced and disappeared when stimulus or response attributes of the two tasks did not overlap (Mayr, 2001; see also Eppinger et al., 2007; but see Kray et al., 2005). Mayr (2001) concluded that older adults' problems in task switching arise from a deficit in differentiating between partially overlapping task-set representations, that is, from resolving task interference. As a consequence of this overlap, older adults show the tendency to update internal control settings all the time when being in a switch situation, i.e., on switch as well as on repetition trials. Therefore, age differences are much larger for mixing than for switching costs.

Summary: behavioral findings
Taken together, healthy elderly exhibit increased mixing costs, while there are no or only small increases in switching costs, compared to young subjects. In elderly, both costs are strongly modulated by memory demands, the availability of cues, and interference by conflicting stimulus-response sets. Older subjects appear to profit from advance preparation to a similar degree as younger subjects which suggests that age-related differences in mixing costs are not due to differences in advance preparation. ERPs can shed more light on this possibility.

ERP studies on age differences in task switching
Neurophysiological measures like event-related potentials (ERPs) can further our understanding of the processes contributing to task switching. Because of their high temporal resolution, they provide an online-measure of information processing in different phases of task performance. This way, they allow distinguishing processing stages that are related to the cue, the target, as well as the response. They make it possible to identify the differential contribution of these processes to task and response preparation, execution and monitoring, and to determine the process-specific impairments in the elderly. ERP studies examining age effects in task-switching are summarized in Table 1. Fig. 1 illustrates an example of a prototypical trial in a taskswitching paradigm in which a cue-, target-, response-, and feedback phase can be separated including the ERP correlates of the different phases discussed in the present review.

Task preparation (cue-locked ERPs)
Fig. 2 shows schematically the cue-locked ERPs in single task blocks, non-switch and switch trials separately for younger and older adults.
2.1.1. Task-set updating (switch positivity / Cue-P3) In cued task-switching paradigms, ERPs in the cue-target interval enable the examination of proactive control processes reflecting preparation for an upcoming task. An ERP component typically elicited by the presentation of a cue signaling a task repetition or switch is a P3like positive deflection that is maximal over parietal scalp regions and peaks approximately 400-600 ms after cue presentation. This cue-related P3 has also been labeled sustained posterior positivity (spP; Kopp et al., 2014) or switch positivity (Karayanidis et al., 2003(Karayanidis et al., , 2011; see also Karayanidis and Jamadar, 2014 for a comprehensive discussion of this point), because it is substantially larger in mixed-than single-task blocks and usually larger for switch as compared to non-switch trials (e.g., Kieffaber and Hetrick, 2005;Lavric et al., 2008;Nicholson et al., 2005;Swainson et al., 2006). However, as the component occurs not only on switch trials but also on repetition trials, the term switch positivity may be misleading. Moreover, this component has a number of properties similar to the classic P300 or P3b interpreted in terms of surprise and context updating (Donchin and Coles, 1988). Therefore, for the sake of consistency with previous literature and functional discriminability between this positivity after cues and the similar P3 component occurring after targets (see Section 3.2.4), we prefer the more neutral terms cue-P3, for the parietal positivity during advance task preparation, and target-P3, for the cento-parietal positive wave during target processing.
Consistent with the view of the classic P300 / P3b account as reflecting context updating in working memory after unexpected or surprising events (Donchin and Coles, 1988;Polich, 2007), the cue-P3 likely reflects the updating of relevant task sets. This assumption is supported by the finding that the size of this positivity is inversely related to the size of the behavioral switch costs, suggesting larger amplitude associated with lower switch costs (Elchlepp et al., 2012;Karayanidis et al., 2009Karayanidis et al., , 2011Kieffaber and Hetrick, 2005;Lavric et al., 2008). However, presentation of dummy cues indicating no specific task set in a full memory-based switching paradigm does not produce this typical cue-P3 switch effect (Gajewski and Falkenstein, 2015a;Gajewski et al., 2010a as the time point of task-set retrieval and updating is individually different and smears out the resulting cue-positivity. Lack of the cue-P3 can also be observed in singletask blocks, in which the cue is redundant and no task updating is required. This suggests that only cues conveying task relevant information elicit a cue-P3 after averaging (Fig. 2).
Comparison of single-and mixed-task blocks yields consistent age differences in the latency of the cue-P3 (Eppinger et al., 2007;Kray et al., 2005;West, 2004;West and Travers, 2008;Karayanidis et al., 2011; but see Gaál and Czigler, 2015). For example, Kray and colleagues (2005) compared mixed-and single-task blocks in younger (mean age = 21.7 years) and older (mean age = 62.9 years) adults using a color word Stroop task. The participants were instructed by a cue to either perform the word task (requiring responding to the word meaning), or the color task (requiring responding to the word color). The results showed a cue-P3 that was larger in a mixed-task block than in a single-task block for both age groups. There were no age effects for cue-P3 amplitude, suggesting that younger and older adults both used the cues for updating a representation of the relevant task set. However, cue-P3 latency was significantly delayed in older adults under switching conditions, indexing that this updating process is slowed in older adults (Kray et al., 2005). Similar findings have been reported by other studies using the cued color word Stroop paradigm (Eppinger et al., 2007;West, 2004;West and Travers, 2008), and also in other tasks like the Wisconsin Card Sorting Test (WCST; Adrover-Roig and Barceló, 2010) or cued letter vs. number classification task (Karayanidis et al., 2011). Additionally, when the duration of the time interval between cue and target is varied, younger adults show no latency differences in the cue-P3, whereas for older adults the component is prolonged when the preparation time is longer (Czernochowski, 2011). Together with the finding of reduced mixing costs with longer preparation time, this indicates that age effects emerge when time pressure is high and that older adults can effectively use the additional time to keep their performance high.
While no age effects in cue-P3 amplitude occur when comparing single-and mixed-task blocks, age effects have been found when switch and non-switch trials within mixed-task blocks are compared. Here, it has been found that the cue-P3 is larger on switch than non-switch trials in younger adults, while this effect is attenuated in older adults (West and Moore, 2005;Karayanidis et al., 2011;Kopp et al., 2014, see Fig. 2). These findings are in line with the assumption that older adults tend to update task sets on each trial when being in a switching situation, whether it is necessary (switch trials) or not (repeat trials) (Eppinger et al., 2007;Friedman et al., 2008;Karayanidis et al., 2011;Mayr, 2001;Whitson et al., 2014).
In their study, Gaál and Czigler (2015) demonstrated that the updating process reflected in the cue-P3 indexes a deliberate preparation strategy. They used a letter-parity task-switching paradigm that either contained an informative cue (informative as for which task should be executed) or a non-informative cue (a fixation cross only carrying information about the timing of the target) and showed that for younger adults the cue-P3 was much larger in conditions in which the cue actually carried information that could be used for task preparation. Older • an anterior shift in target-P3 and attenuated fronto-central slow negativities in single-task and mixed-task blocks for older adults Czernochowski (2011) 20-25 years and 61-83 years; n = 40 switching between digit number and digit identity task; alternating runs paradigm with task switches every after 0, 1, or 2 repeat trials • mixing costs slightly larger for older than younger adults • short preparation times (600 ms): smaller frontal slow waves associated with pronounced congruency costs in older adults; centro-parietal P3 smaller for older than young adults • late sustained positivity present in young subjects (preswitch > switch) but absent in elderly over the last 400-500ms Eppinger et al. (2007) 21.3 years (SD = 1.8) and 63.7 years (SD = 2.6); n = 24 cued switching between color and word naming in a Stroop task; ratio of conflict trials (incompatible vs. compatible Stroop trials) was manipulated • cue interval: cue-P3 varied as a function of conflict ratio and a later positive component was larger for switch than repeat trials • target interval: a negativity after incompatible trials (Ni) was delayed for older adults; the Ni also varied as a function of conflict ratio and was larger for switch than for non-switch trials • response-locked ERPs: correct response negativity (CRN) larger on incompatible trials and for younger adults; this compatibility effect was enhanced when incompatible trials were infrequent Friedman et al. (2008) 23.1 years (SD = 2.8) and 71.0 years (SD = 6.9); n = 40 cued switching between digit value (more vs. less than 5) and digit parity (odd vs. even) task; task cue and target were presented simultaneously • trials in single blocks had faster RTs than pre-switch trials occurring during blocks when a switch was expected (mixing costs), relative to pre-switch trials, switch trials showed longer RTs (switching costs), and relative to preswitch trials, post-switch trials showed continued RT prolongation (post-switching costs) • age-effects were found for switching and post-switching costs • target-locked ERPs: fronto-central P3 and sustained positivity for switch trials in young adults, frontal P3 in switch, noswitch, and pre-switch trials (anterior shift) and sustained negativity for post-switch trials in older adults • response-locked ERPs: the amplitude of a medial frontal negativity (MFN) increased as a function of executive demands in ERPs of young but not older adults (noswitch < pre-switch < post-switch < switch trials) Gaál and Czigler (2015) 18-25 years and 60-75 years; only women; n = 79 cued switching between letter (vowel vs. consonant) and digit parity (odd vs. even) task; paradigm either contained informative cue, informative cue and nogo stimuli, or non-informative cue and nogo stimuli • younger adults: shorter RTs and larger cue-P3 in repeat than switch trials for task without nogo stimuli, but similar RTs and cue-P3 in repeat and switch trials in tasks with nogo stimuli • older adults: lack of cue-P3, CNV, and target-P3 in tasks with nogo stimuli • no age differences in cue-P3 in informative cue condition (without nogo stimuli) Gaál and Czigler (2017) 18-25 years and 60-75 years; only women; randomly divided in task-switching training and passive control group; n = 79 cued switching between letter (vowel vs. consonant) and digit parity (odd vs. even) task or between color (green, orange, blue); paradigm contained informative cue and nogo stimuli • mixing costs in the older adults decreased with training and became similar to that observed in young adults, no changes for switching costs • only young adults had cue-P3 before training which did not change with training; older adults cue-P3 emerged with training • target-locked N2 increased with training for older adults only • only young adults had target-P3 before training which also increased with training; older adults target-P3 emerged with training but did not reach the size of young adults' target-P3 • training led to improvement in alerting and orienting networks, and in other variants of task switching paradigms • behavioral and ERP changes were preserved after one year (continued on next page) P.D. Gajewski et al. Neuroscience and Biobehavioral Reviews 92 (2018) 255-275  Memory-based switching between digit value (more vs. less than 5), digit parity (odd vs. even), and font-size (small vs. large) task; alternating runs with task switches every 3trials; informative cues were presented in the single blocks, uninformative ones (XXX) in mixed blocks • active participants had lower mixing costs (RT and accuracy) and lower switching costs (accuracy) than low active ones; costs were negatively correlated with selfreported level of physical activity • cue-locked frontal CNV smaller in active than low active group • in target-locked ERPs, active individuals showed an earlier P2, larger fronto-central N2, and smaller target-P3 in switch than non-switch trials, no difference in the target-P3 in the sedentary group Gajewski et al. (2010a) 18-30 years and 48-58 years, assembly line and non-assembly line workers, n = 91 cue-and memory-based switching between digit value (more vs. less than 5), digit parity (odd vs. even), and fontsize (small vs. large) task; cue and target presented simultaneously (CSI = 0 ms), uninformative cues (XXX) in memory-based blocks  Gajewski et al. (2017a) 46.5 years (SD = 4.5); n = 57; assembly line workers; divided into a cognitive training group (CT) and waiting control group Memory-based switching between digit value (more vs. less than 5), digit parity (odd vs. even), and font-size (small vs. large) task; alternating runs paradigm with task switches every 3rd trials uninformative cues (XXX) in memory-based blocks • reduction of mixing costs in accuracy after cognitive training (CT), stable effects in the follow-up measure • reduction of mixing costs in accuracy in the waiting control group after CT.
• higher amplitudes of N2 in target-locked ERPs and larger Ne /

ERN after CT in both groups
• Increase of the target-P3 after CT, and a decrease of target-P3 after follow-up measure Gajewski et al. (2017b) 30-60 years; (43.9); n = 76; divided in high and low burnout scores and high and low depression scores; Memory-based switching between digit value (more vs. less than 5), digit parity (odd vs. even; alternating runs paradigm with task switches every 2nd trials • no differences in switching costs • lower CNV, P3a and P3b in burnout • larger Ne / ERN and FRN in burnout (Gajewski et al., 2017c) Goffaux et al. (2008) 24.5 years (SD = 3.4) and 75.8 years (SD = 4.4); n = 47 older adults were split into high-and lowworking memory groups; cued task switching with three semantic classification tasks (living vs. nonliving, large vs. small, wide vs. narrow); cue present throughout the cue-target interval • older adults with poor working memory had larger RT mixing costs than younger adults, while all groups had similar RT switching costs • older adults with poor working memory showed negative slow wave after cues in single blocks while younger adults did not • older adults with good WM showed larger negative slow waves especially at fronto-lateral sites than younger adults in repeat and switch trials Hillman et al. (2006) 19.4 (SD = .3), 19.4 years (SD = .2), 63.7 years(SD = .9), and 65.9 years (SD = .8); n = 66 divided in physically active and sedentary groups cued switching between digit value (more vs. less than 5), digit parity (odd vs. even) task; cue and target presented simultaneously • increasing age was associated with three latend variables, namely slower non-decision processes, slower rate of evidence accumulation about the target, and higher response criterion (amount of evidence required for a decision) • age effects on mixing costs were evident only on response criterion, whereas age effects on switch cost were present for all three latent variables • cue-P3 smaller and more frontally distributed for older adults • pre-target negativity increased with age for switch and repeat trials • mixing and switching effects on target-P3 became larger and more frontally distributed with age Kopp et al. (2014) 18-39 years and 64-83 years; n = 40 switching between sorting rules in card sorting task; feedback served as cue for next trial • age-related reduction of cue-P3 after switch cues • age-related frontal shift of target-P3 (continued on next page) P.D. Gajewski et al. Neuroscience and Biobehavioral Reviews 92 (2018) 255-275 Küper et al. (2017) 65-87 years; n = 114; randomly assigned to a cognitive training, passive and active control group cued switching between color and word naming in a Stroop task; task switching was administered in preand post-test; (cognitive) training did not involve taskswitching • cognitive training group only showed increase in response accuracy at posttest, irrespective of task and trial type; no training-related effects in reaction times • cognitive training associated with increase in cue-related N2 amplitude and decrease of P2 latency in single blocks • cognitive training also associated with an amplitude decrease in target-P3 • increase in the cue-P3 emerged after both cognitive and relaxation training Schapkin et al. (2014) 21-35 years and 51-63 years; n = 93 switching between digit value (more vs. less than 5), digit parity (odd vs. even), and font-size (small vs. large) task; alternating runs paradigm with task switches every 3 trials; informative (cue-based) or uninformative (memory-based) cues were used in mixed blocks • larger mixing costs for older than younger adults in RT and accuracy during memory-based switching • presence of cues reduced age differences in mixing costs for accuracy but not RT • larger RT switching costs in younger than older adults during memory-based switching • CNV reduced during memory-based switching and correlated with accuracy mixing costs in older adults • target-locked occipital N1 and fronto-central P2 larger in older than younger adults; P2 latency correlated with RT mixing costs in older adults • delayed N2 and delayed and reduced P3b in older adults Themanson et al. (2006) 18-21 years and 60-71 years; n = 66; divided in high and low physically active cued switching between digit value (more vs. less than 5), digit parity (odd vs. even) task; cue and target presented simultaneously • older adults exhibited slower RTs in mixed-task blocks and smaller ERN amplitude compared to younger adults • smaller mixing costs for physically active older adults • decreased Ne / ERN and increased post-error slowing for older and younger physically active participants Tsai and Wang (2015) 60-77years; n = 64; divided in open-skill, closed-skill, and sedentary-behavior (control) group cued switching between digit value (more vs. less than 5), digit parity (odd vs. even) task; cue and target presented simultaneously • the two exercise groups exhibited shorter RTs and larger P2 and P3 amplitudes across all conditions • smaller switching costs, faster RTs and larger P3 in the switch condition for open-skill as compared with closed-skill and control group West (2004) 21.36 years and 72.21 years; n = 28 cued switching between color and word naming in a Stroop task • cue-locked: occipital-parietal N1 larger for mixed than single task blocks for younger but not older adults; cue-P3 elicited for mixed but not single blocks with longer latency for older adults; occipital-parietal negative/ anterior frontal positive slow wave larger for mixed-than singletask blocks; effect attenuated over occipital regions in late epoch for older • target-locked: N450 for younger present in single-and mixedtask blocks, for older adults absent in single blocks; SP larger for younger in color condition; a second sustained potential observed for older adults for color incongruent vs. congruent trials West and Moore (2005) 21.36 years and 75.21 years; n = 24 cued switching between color and word naming in a Stroop task • no switching costs and no age effects in behavioral data • cue-locked: switch trials elicited enhanced Cue-P3 in younger adults, which was attenuated in older adults; switch trials elicited enhanced left frontal slow wave in younger which was attenuated in older adults West and Travers (2008) 20.75 years and 70.58 years; n = 24 cued task switching between colour and word naming Stroop task; information reduction paradigm with two cues for each task • no reliable age effects in RT mixing costs • age-related frontal shift in cue-P3 and reduction of cue-locked fronto-polar positivity related to task mixing • reduction in amplitude of parietal slow wave with cue retrieval (cue vs. task repetitions) • sustained activity over frontal and occipital-parietal regions after cue onset was similar for older and younger adults in the word task but more strongly expressed for the older in the color task (task repetitions vs. alternations) • enhanced parietal target-P3 in single-vs. mixed-task blocks in younger adults only • sustained modulations of ERPs in single-vs. mixed-task blocks over central-parietal and frontal regions beginning 800 ms after target onset in older but not younger adults (continued on next page) P.D. Gajewski et al. Neuroscience and Biobehavioral Reviews 92 (2018) 255-275 adults needed training in the task to develop the same cue-P3 pattern as was found in younger adults (Gaál and Czigler, 2017). However, this study did not find a larger cue-P3 for switch as compared to repeat trials, which might be partly due to the fact that the exact nature of the preparation process had been altered by the insertion of unpredictable nogo trials into the paradigm (cf. Kleinsorge and Gajewski, 2004; but see Karayanidis and Jamadar, 2014;Whitson et al., 2014). A recent study by Küper et al. (2017) analyzed effects of four months multidomain cognitive training on the switch ability in a combined Stroop-switch task in elderly individuals (65-86 years). They found an enhanced accuracy in non-switch and switch trials accompanied by enhanced cue-P3 after cognitive training, but a similar increase was observed in a relaxation group that did not show any improvements in performance. Specific training induced effects associated with enhanced performance were found in increased negative potentials during target processing in both switch and non-switch trials as discussed below in more detail.
Another common finding concerning age-related changes in the P3 is that it usually shows a parietal topographical distribution in younger adults. In contrast, in older adults cue-P3 topography is more equally distributed over the scalp without a clear parietal focus (Fig. 2) (Eppinger et al., 2007;Karayanidis et al., 2011;Kray et al., 2005). This result has led to the speculation that older adults need to recruit frontal areas to a larger extent than younger adults to keep their performance up (Cabeza et al., 2004;Friedman, 2008;Reuter-Lorenz and Cappell, 2008). This idea is supported by the evidence that this more evenly distributed cue-P3 is especially pronounced under more demanding conditions (like the color task in the color Stroop paradigm) in which general switch costs are also higher for older adults. Additionally, there is evidence showing that preserved executive function in elderly participants with high performance levels is accompanied by enhanced recruitment of prefrontal cortical mechanisms (De Sanctis et al., 2009; but see Adrover-Roig and Barceló, 2010;Nashiro et al., 2018).
Together, findings concerning the cue-P3 in the preparatory interval show that older subjects reveal less efficient updating on switch trials. This is reflected in a reduced amplitude difference between switch and non-switch trials (cf. Fig. 2). At the same time, the ERPs show evidence for an additional recruitment of frontal resources in older subjects, as shown in a more frontal topography of the cue-P3. This suggests that proactive executive control processes during preparation for an upcoming task change in older age to keep performance high and indicates a similar advance preparation in non-switch and switch trials.

Preparatory activity (negative slow waves / CNV)
A second ERP index that is usually present in task-switching studies during the interval between cue and target and in which age differences are usually found are negative going slow waves (Fig. 2). They are sometimes termed CNV (contingent negative variation; Walter et al., 1964;Brunia, 1999;Brunia and van Boxtel, 2001), pre-target negativity (Karayanidis et al., 2011;Karayanidis and Jamadar, 2014), or sustained frontal negativity (sfN; Kopp et al., 2014). These negative slow waves are maximal just before target presentation and show a broad central distribution over the scalp. They are thought to reflect processes of preparation for the upcoming task like task-set retrieval and maintenance until target presentation. In particular, it has been assumed that the amplitude reflects the extent to which informative cues are utilized. These negative slow waves involve an anticipatory aspect, indexing expectation of an upcoming target-onset, as well as a motor aspect linked to preparedness to respond to the target (Brunia and van Boxtel, 2001). Also, they are often larger on task repetition than task-switch trials (Goffaux et al., 2006;Kieffaber and Hetrick, 2005;Nicholson et al., 2005, but see Rushworth et al., 2002). This can be interpreted as an active maintenance of the same task set across two or more consecutive trials or alternatively as a less efficient preparatory activity after a task switch.
Age differences in slow wave amplitudes can be found mainly in mixing conditions (Schapkin et al., 2014). For example, in the above described study by Kray et al. (2005), older adults generated substantially larger negative slow waves under switch than non-switch conditions, while no reliable differences were found for younger adults (for similar results, see Goffaux et al., 2008;Karayanidis et al., 2011). Because an enhanced negative slow wave has been previously found on trials associated with particular effort (Falkenstein et al., 2003), it can be concluded that older adults have problems in maintaining the relevant task set under the more demanding switching conditions, and hence, invest more effort to do so. In contrast, West (2004) found larger negative slow waves under switch as compared to non-switch conditions for younger and older adults in an early time window, while in a later time window this difference was present only for younger adults. The differences in result patterns between the two studies seem contradictory at first glance, however, they could be due to differences between the study designs. For example, it is known that slow wave activity is modulated by task difficulty. They are more pronounced when more difficult tasks are used (e.g., incongruent Stroop stimuli; Kray et al., 2005;West, 2004) and they are less apparent when less demanding tasks are used (e.g., classification tasks; Kieffaber and Hetrick, 2005). Although both above studies applied a cued color word Stroop paradigm, the task version used by West (2004) probably created much more time pressure and confronted participants with a much less practiced task (one color-to-key practice block, one practice block for each task (word single, color single, mixed) as compared to the study by Kray et al. (2005) with two and four practice blocks. Also, the participants in the study by West (2004) were approximately ten years older than those in the study by Kray et al. (2005). Finally, both studies used different reference positions, which may affect distribution and morphology of potentials (Picton et al., 2000). With this in mind, the seemingly contradictory result can be reconciled: In easier conditions, • switching costs also associated with delay in response selection • age-related increase in mixing costs associated with greater interference at level of decision-response mapping and response programming for repeat trials in mixed-task blocks P.D. Gajewski et al. Neuroscience and Biobehavioral Reviews 92 (2018) 255-275 older adults might still be able to maintain the task sets by employing maintenance processes to a larger extent than younger adults, while in more demanding conditions in which even younger adults have to invest more in maintenance, older adults are impaired and fail to maintain the task sets. This idea is supported by data from West and Moore (2005) demonstrating slow wave differences between switch and non-switch conditions of a color word Stroop task. This difference was larger for the more demanding color-naming task and was attenuated in older adults. Interestingly, this failure to activate the neural generators of the frontal slow wave in older adults went hand in hand with an increase in the number of intrusion errors on color incongruent trials in the mixed task condition (West and Moore, 2005). In a similar vein, Adrover-Roig and Barceló (2010) found comparatively larger mixing costs together with an attenuated frontocentral negative slow wave in middle aged and older adults, who scored low on tests of executive functioning. A similar age-related effect was found by Sterr and Dean (2008) and Schapkin et al. (2014). Further, middle-aged employees with subclinical burnout (i.e. employees reporting symptoms of burnout without a clinical diagnosis) showed a trend for lower performance and a reduced negative slow wave compared to healthy controls in a memorybased task switching (Gajewski et al., 2017b).
The interpretation of the age-related reduction negative slow wave as reflecting impairment of older adults in maintenance processes is intuitive, given the fact that in most task-switching studies the cue disappears some time before the target is presented, and thus, the information related to the cue needs to be maintained. However, this view had to be modified on the basis of studies showing that negative slow waves still occur in task-switching paradigms in which the cues are visible throughout the cue-target interval (CTI) (Nicholson et al., 2005). Fig. 1. A schematic illustration of a trial in a task-switching paradigm and the corresponding ERP and their topography in the cue, target, response, and feedback phase. The ERP reflects a grand average in the task-switch condition with a subsequent error response. The relevant components are labelled.
Negativity is plotted upward. Unpublished data.

Fig. 2.
An example of cue-locked ERP for single task (black), task repetition (green) and switch trials (red) in young (left), and old individuals (right). The relevant components are labelled. Negativity is plotted upward. Unpublished data (For interpretation of the references to color int his figure legend, the reader is referred to the web version of this article).
These results strongly argue against negative slow waves exclusively reflecting maintenance processes of a task set. One explanation could be that the negative slow waves also include processes of target expectation or response-set preparation, as has been previously suggested for the CNV (Brunia and van Boxtel, 2001;Falkenstein et al., 2003;Hohnsbein et al., 1998): the closer the imperative stimulus, the stronger the activation of response-related preparation. Yet, it should be noted that a specific response is usually not indicated by a cue and only a stimulus-response set (task-set, e.g., odd-even) can be prepared.
A similar explanation of the negative slow wave has been put forward by Goffaux and colleagues (2008). They proposed that negative slow waves might be an index of task-set retrieval processes that need to be recruited irrespective of the duration of cue presentation. To disentangle whether age differences in negative slow waves are actually due to older adults' problems in task-set maintenance or retrieval, Goffaux et al. (2008) conducted a study in which the cue was present until shortly before target presentation to minimize maintenance demands. They found that older but not younger adults showed slow wave activity in single blocks. This effect was most pronounced for older participants with low working-memory performance. Additionally, older participants scoring high on working-memory measures performed the task as well as younger adults and had larger frontal negative slow waves for repeat and switch trials. Since slow wave activity was found despite the continued availability of the task cue, these findings were interpreted as preparation for the upcoming task by retrieving task-relevant attributes (Johansson and Mecklinger, 2003;Goffaux et al., 2006). Older adults with high scores on working-memory measures kept their performance in the more difficult mixed-task condition up by exerting more executive control as indexed by greater frontolateral activation. In contrast, older adults with low working memory needed to exert trial-to-trial preparation even in single-task contexts. Also, the study by Gajewski et al. (2010a) showed smaller slow waves in older low performers than in older high performers or younger subjects in a memory-based mixing block corroborating the findings by Goffaux et al.(2008). Additionally, the slow-wave amplitude was highly correlated with task switch performance. These considerations are also in line with a recent study by Schapkin et al. (2014). Although this study did not find differences in slow waves between switch and non-switch conditions, clear age differences occurred: larger slow waves were observed in younger as compared to older participants both in cue-and memory-based mixed blocks relative to single task blocks (see also Gaál and Czigler, 2015).
Moreover, slow wave activity is not only modulated by task difficulty and working-memory capacity, but also by short-term effort, preparation efficiency, and fatigue. Increased negative slow wave activity is associated with short-term mobilization of cognitive resources in speeded tasks (Falkenstein et al., 2003;Hohnsbein et al., 1998), i.e., a motivational aspect of preparation and efficiency of task-set preparation with stepwise amplitude reduction by increased mental fatigue during task switching (Lorist et al., 2000). Finally, regular physical activity in seniors is associated with lower slow wave amplitude and improved performance during task switching (Gajewski and Falkenstein, 2015a) and other paradigms like the Sternberg task (Kamijo et al., 2010) or the S1-S2-S3 paradigm (Hillman et al., 2002), consisting of a visual warning stimulus (S1), a discrimination task (S2) and a stimulus indicating the time point for responding (S3). This suggests that slow wave amplitude can be increased by short-term recruitment of processing resources and decreased by lower use of preparatory resources due to more efficient neuronal processing.

Summary: ERPs during advance task preparation
While most of the behavioral studies did not find evidence that preparation is generally reduced by age, the ERP results showed clear differences between older and younger subjects. The cue-P3 is larger on task switch than non-switch trials in younger adults while this difference is attenuated in older age, suggesting that older individuals update task sets in non-switch as well as in switch trials or fail to update task sets on switch trials. Additionally, the cue-P3 has a more frontal focus in older than in younger subjects. The subsequent negative slow wave has often been found to be larger on non-switch compared to switch trials in younger subjects, whereas older participants show larger amplitudes on task-switch trials. This suggests a stronger maintenance and preparation of task sets in elderly to maintain a reasonable level of performance. Thus, on the behavioral level older adults may show no impairments because their inefficient cue processing is compensated by an additional recruitment of prefrontal cortical mechanisms and enhanced maintenance of a task-set on switch trials. Hence, performance and ERP data during task preparation suggest that task-switching behavior of older subjects is not impaired because of compensatory mechanisms during advanced task preparation, despite some alterations in task-set management. Consequently, the larger mixing costs in the elderly probably result from effects in the post-target interval that cannot be compensated sufficiently. Therefore, in the next chapter, aging effects in targetlocked ERPs will be examined.

Task implementation (target-locked ERP)
After the target onset, task-set retrieval and action-related processes are required. In particular, they are related to attention allocation to the stimulus, stimulus encoding, retrieval of a currently relevant S-R mapping, implementation of this mapping (i.e., response selection), and finally execution of the selected response. Each of these processing steps has been associated with a particular ERP component. As outlined below in more detail, most of them differ between younger and older adults and between non-switch and switch trials. Fig. 3 shows an example of findings in the target-locked ERP in single task blocks, non-switch and switch trials, separately for younger and older adults and illustrates task-and age-related differences in the P2, N2 and P3b components sensitive to task switch vs. non-switch and single task blocks described below in more detail.

Selective attention (P1 and N1)
The bilateral, parieto-occipital components P1 and N1 are thought to reflect the arrival of sensory information in the midbrain and both are modulated by attentional requirements of the task (Hillyard and Anllo-Vento, 1998;Mangun and Hillyard, 1995). It has been reported that P1 and/or N1 are larger in older than younger adults (Yordanova et al., 2004b) but their latency seems to be only occasionally affected by age (see De Sanctis et al., 2008;Friedman, 2008 for overviews). This amplitude increase was interpreted as deficient prefrontal inhibitory control over thalamically-mediated gating of inputs to sensory cortex (Alho et al., 1994;Knight, 1991) or rather enhanced attention as compensatory strategy (Hillyard and Anllo-Vento, 1998;Yordanova et al., 2004b). The N1 enhancement in older age was also confirmed in task-switching studies (De Sanctis et al., 2009;Schapkin et al., 2014) but to date no differences between switch and non-switch conditions have been reported.

Stimulus-response retrieval (P2)
The second ERP-component increasingly reported in the task-switching context is the anterior P2 with a peak about 200 ms after stimulus onset. Several authors related the P2 to task-set activation (Kieffaber and Hetrick, 2005; Adrover-Roig and Barceló, 2010;Finke et al., 2011Finke et al., , 2012, and interpreted this component as an index of retrieval of S-R associations or evaluation of salience and relevance of action relevant stimuli (see also Barceló et al., 2008;Periáñez and Barceló, 2009;Rushworth et al., 2002;Tsai and Wang, 2015). This functional interpretation includes working memory involvement that consists of a network localized in frontal brain areas (Braver et al., 2007;Braver and West, 2008). This is in line with the interpretation of the P2 offered by several authors using other paradigms (e.g., Gajewski et al., 2008Gajewski et al., , 2013 P.D. Gajewski et al. Neuroscience and Biobehavioral Reviews 92 (2018) 255-275 The existing literature indicates that the P2 is the first target-locked component that differs across non-switch and switch trials (Fig. 3). However, the switch effects on the P2 are inconsistent in the literature. Kieffaber and Hetrick (2005) reported smaller target-locked P2 on switch than non-switch trials and a significant positive correlation between P2 amplitude and switch costs, suggesting lower performance when the P2 was larger. In contrast, Adrover-Roig and Barceló (2010) reported no difference between non-switch and switch trials in the target-locked P2. As to aging effects in the P2 in switch tasks, Schapkin et al. (2014) also found a larger and delayed target-P2 in mixed blocks than in single task blocks as well as a generally larger P2 in older than younger participants. No amplitude difference between switch and nonswitch trials was found. Additionally, P2 latency was positively correlated with RTs and mixing costs in older participants. In this context, it is worthwhile to emphasize that Adrover-Roig and Barceló (2010) found a positive correlation between restart costs and the frontal cue-P2 amplitude, suggesting that higher restart interference is associated with a larger frontocentral P2. Gajewski and Falkenstein (2015a) found a shorter P2 latency accompanied by lower mixing costs in physically fit vs. unfit seniors. Tsai and Wang (2015) reported a larger P2 and smaller local switch costs in physically fit compared to sedentary elderly. Additionally, they reported a gradual decrease of the P2 amplitude from task switch, across non-switch, pre-switch, and post-switch trials. In a recent training study with middle-aged participants task-switch performance was enhanced, the P2 latency for switch trials was shortened and the amplitude reduced after cognitive training (CT), whereas these effects were not observed in a waiting control group (Fig. 4 bottom). This latter group served as a no-contact control group until a post-measure (t2) was conducted. Afterwards, this group also was trained for three months and the results evaluated in a follow-up measure (t3). The same ERP and behavioral effects occurred in the waiting control group after cognitive training that validated this observation (Gajewski et al., 2017a,; cf. also Küper et al., 2017).
These findings suggest that the P2 is the first ERP in the targetlocked data related to performance in task switching. It seems to be modulated by age but the studies are scarce and the results inconsistent.
Nevertheless, its amplitude and latency were positively correlated with mixing and switching costs. This suggests that the larger the P2 amplitude and the longer the P2 latency, the lower the switch performance. Accordingly, P2 amplitude is larger and its latency delayed in older vs. younger adults. Both P2 amplitude and latency can be reduced by physical and cognitive training that was associated with enhanced performance. These results indicate that frontal P2 is related to the retrieval of S-R associations from working memory to select the correct response. This process seems to be diminished in older age and can be restored by training.

Conflict processing during response selection (N2)
Following the P2, a large negative peak occurs, the N2 (Fig. 3). In task-switching studies, the N2 usually peaks between 280 and 340 ms and shows a frontocentral distribution. The source of the N2 has been found in the rostral zone of the anterior cingulate cortex (ACC) at the border between Brodman area 6 and 8 (Ullsperger and von Cramon, 2004). It has been associated with categorization and decision making, mismatch, generation of motor behavior, and translation of intentions to actions (Ritter et al., 1979(Ritter et al., , 1982Turken and Swick, 1999; for reviews, see Folstein and van Petten, 2008;Hämmerer et al., 2014;Paus, 2001;Pires et al., 2014). Since the N2 is usually larger in conflict conditions, it has also been related to response conflict processing (Kopp et al., 1996;van Veen and Carter, 2002;Yeung and Cohen, 2006) or to a more general mechanism of response selection, i.e., implementation of S-R mappings, which is more intense under conflict conditions (Berchicci et al., 2017;Gajewski et al., 2008Gajewski et al., , 2010bKarch et al., 2010;Yeung and Cohen, 2006). These different interpretations of the N2 are not mutually exclusive: due to the mismatch or conflict resolution, the selection of the correct response is more demanding as indicated by the larger, broader, and often delayed N2. Consequently, the N2 is usually larger and/or delayed on task-switch than non-switch trials (Barceló et al., 2000;Finke et al., 2012;Gaál and Czigler, 2015;Gajewski et al., 2010b;Gehring et al., 2003;Goffaux et al., 2006Goffaux et al., , 2008Jackson et al., 2001Jackson et al., , 2004  3. An example of target-locked ERP for single task (black), task repetition (green) and switch trials (red) in young (left), and old individuals (right). The relevant components are labelled. Negativity is plotted upward. Unpublished data (For interpretation of the references to color int his figure legend, the reader is referred to the web version of this article). , 2006Poulsen et al., 2005;Rushworth et al., 2002;Wylie et al., 2003). This difference between switch and non-switch N2 suggests that during implementation of task sets (i.e., S-R sets), interference from the currently irrelevant S-R set has to be overcome (inhibited), which is a time and resource consuming process (Allport et al., 1994;Allport and Wylie, 2000). In more mechanistic terms, the process reflected in the N2 is increased and delayed due to a longer activation time of the correct response until a selection threshold is exceeded on switch compared to repetition trials (Gajewski et al., 2010b;Koch and Philipp, 2005;Schuch and Koch, 2003). In line with this interpretation, Karayanidis et al. (2011) and Whitson et al. (2014) found that effects on mixing and switching costs were evident on response criterion, i.e., the amount of evidence required to trigger a decision (see below). The involvement of the N2 in response-related processes has recently also been documented by Berchicci et al. (2016) who found the same N2 amplitude in stimulus-as well as response-locked analysis.
However, paradoxically the N2 in single-task blocks is even larger than that on switch or non-switch trials of mixed blocks (cf. Fig. 3) (see also Falkenstein, 2012, 2015a;Kray et al., 2005;Karayanidis et al., 2011;Schapkin et al., 2014). This seems counterintuitive at first glance, but if one assumes that the N2 reflects a response-selection process, a larger and earlier N2 which covaries with shorter RTs may suggest lower response selection variability. In contrast, a smaller and broader N2 may indicate larger latency jitter of the selection process that leads to increased variability of RTs and even enhanced error rates (cf. Hohnsbein et al., 1998). This supports the pattern of larger and earlier N2 in single-task blocks in which response selection is not affected by interference due to overlapping S-R sets Karayanidis et al., 2011;Kray et al., 2005;Schapkin et al., 2014). However, this account does not explain the larger N2 on switch than non-switch trials, as the RTs and RT-variability are higher in the former than the latter. Thus, it is quite possible that an additional process required for the task switch (for example inhibition of the irrelevant task set) is reflected in the increase of the N2 amplitude relative to non-switch trials. This increase may be due to partial overlap in time of the "pure" N2 observed in single-task blocks with a conflict-or inhibition-related process on switch or non-switch trials in mixing blocks. Further research is necessary to clarify this issue.
The N2 is generally reduced and delayed with increasing age (for reviews, see Friedman, 2008;Hämmerer et al., 2014;Pires et al., 2014). An age-related delay and reduction of the target-locked frontocentral N2 has also been found in other paradigms and suggests a decline in control over competing S-R mappings in older age, resulting in decreased performance ( 4. 4a: Effects of cognitive training (CT) on the P2 and N2 in middle-aged participants (mean 46.4 years old). The ERP in non-switch (top) and task switch trials (bottom) for the three time points (t1: pre-measure (black); t2: post-measure (red); t3: follow-up-measure (green)) and two groups (left: CT: cognitive training between t1 and t2 and a follow-up measure at t3, and right: Control + CT: a waiting control group between t1 and t2 followed by CT between t2 and t3). From: Gajewski et al. (2017a). 4b: Effects of CT on the N2 in old participants (mean 70 years old) for single blocks (black), nonswitch (green) and switch trials (red) for two time points (t1: pre-measure (dotted); t2: post-measure (solid)) and two groups (left: CT: cognitive training, right control group). From:  (For interpretation of the references to color int his figure legend, the reader is referred to the web version of this article).
There are to date only a few studies explicitly contrasting the frontocentral N2 in younger and older individuals in the task-switching paradigm. In a correlational study, Karayanidis et al. (2011) analyzed ERPs from a classic task-switching paradigm (Rogers and Monsell, 1995) across different age groups between 18 and 80 years. They found a larger N2 on switch trials compared to non-switch trials across all groups. This N2 enhancement on switch vs. non-switch trials was larger in the oldest age group (their Fig.7). However, as the authors note, this effect may also be due to an earlier onset of the switch effect on the P3 in older subjects (see below) rather than being a genuine N2 effect. Alternatively, it may be interpreted as age-related increase of interference that has to be resolved on switch trials. In contrast, Gaál and Czigler (2015) showed a switch effect with larger N2 on switch compared to non-switch trials in younger but not in older adults after informative task cues, as well as a generally larger N2 in young vs. old individuals. After non-informative cues, the N2 was considerably smaller and no age or switch effects were seen. The study by Schapkin et al. (2014) compared the frontocentral N2 in a single-task block, and cue-and memory-based mixed blocks between younger and middleaged groups. They observed a delayed N2 in the elderly participants, whereas no age-related amplitude difference was found. Additionally, the N2 latency correlated with mixing costs in accuracy both in younger and middle-aged adults.
A number of studies analyzed a functionally similar negative component with frontocentral distribution but longer latency, the so-called N450, elicited by incongruent color-naming trials in the combined task switching and Stroop paradigm. For example, West (2004) found no task-switching effects on the N450 in young participants, whereas a lower N450 was found in the older as compared to the younger group in the mixed cued-block, when conflict was greatest. This is in accordance with behavioral data showing lower performance in older than younger participants in the conflict condition. In keeping with this finding, Gajewski et al. (2015aGajewski et al. ( , 2015b found both a reduced N2 in a switch task and a N450 in a Stroop task in low compared to high physically active elderly performers (see next section). Kray et al. (2005) asked younger and older participants to switch among word reading and color naming tasks. They obtained a negative wave on incompatible trials, the so-called "Ni", closely corresponding to the N450. Larger Ni amplitude was observed on incompatible vs. compatible trials in both age groups while a delayed Ni latency was observed in the high conflict condition in the older group only, suggesting a decline in conflict processing. In a follow-up study, Eppinger and colleagues (2007) analyzed age effects using almost the same paradigm as in the former study, but they additionally varied the conflict context by manipulating the ratio of compatible and incompatible Stroop trials between blocks of trials. This should modulate conflict requirements compared to the study of Kray et al. (2005). The results showed no age differences in mixing costs, but older participants showed larger interference effects when the ratio of incompatible trials was low and conflict less expected. They also found a larger Ni for switch compared to non-switch trials as well as a later and more negative Ni for older than younger individuals. Importantly, the authors reported a larger Ni on conflict trials in the older but not in the younger age group in a later part of the Ni, suggesting more control during interference processing in the older group.
In a study by Friedman et al. (2008), younger and older participants switched between two numerical tasks in a randomized order without advanced preparation, i.e., the task cue was presented simultaneously with the target. The ratio of switch trials was only 10% to enhance the need for cognitive control. Performance regarding local switch costs (and post-switch costs) was reliably reduced in older than younger adults, whereas no age difference was found in mixing costs. The authors found an enhanced sustained negativity in the time range between 100 and 300 ms for switch trials in old versus young adults. Thus, the N450 / Ni corroborates the pattern found in the N2, showing a general amplitude reduction associated with lower performance with age and larger amplitudes in high vs. low interference trials. This suggests increased executive demands in older age.
Taken together, the frontocentral N2 is a prominent ERP that generally occurs in the task-switching paradigm. Its amplitude decreases with age. The N2 is larger on task switch than repetition trials mainly in older adults. In contrast, in younger individuals this pattern is less consistent. Moreover, the N2 is largest and its latency shortest in singletask blocks as the response selection is not hampered by interference from other tasks. The findings indicate that interference processing before a correct response is selected is compromised in older age. Interindividual differences in task-switch performance and in N2 amplitude and latency increase with age and depend on a number of endogenous and exogenous factors that will be shortly outlined in the following section.
2.2.3.1. Factors affecting the N2 in aging during task switching. Some recent studies investigated diverse endogenous and exogenous factors on executive functions in older age using ERPs during task switching. Most of them focused on the frontocentral N2 to investigate its variability and relationship to task-switch performance. A number of cross-sectional studies showed differences in performance and ERPs between two age-matched groups that differed with regard to a genetic variant or level of physical activity. Other reports showed deficits in performance and reduced ERPs that were ameliorated by training. In some cases, performance and ERPs even reached a similar level to young participants.
For example,  analyzed effects of Val66Met polymorphism of the brain-derived neurotrophic factor (BDNF Val66Met) affecting neuronal integrity, cell growth, and synaptic plasticity, and consequently cognitive functions, particularly executive functions (e.g., Egan et al., 2003;Erickson et al., 2008). A sample of 131 participants older than 65 years was divided into two groups: a homozygote Val/Val group and a combined Val/Met and Met/Met group. Previous studies found that persons with the Met allele have lower cognitive performance than Val allele carriers in executive control tasks. However, this relationship reverses with increasing age as shown using a task switching paradigm in a longitudinal study (Erickson et al., 2008). Indeed, the Met allele group showed generally better performance that is faster RTs and lower intraindividual variability of speed as well as lower error rates in the switch condition in a memory-based switching task compared to the Val/Val group. These superior behavioral effects were accompanied by an enhanced N2 in the switch condition of the memory-based block. The N2 latency was positively correlated with the rate of errors, suggesting that the more delayed the N2, the higher the error rates. Similar effects of the Val66Met polymorphism were observed in respect to the N450 in a Stroop task . In other words, lower neuronal integrity may thus lead to a more variable and delayed response-selection process that in turn reduces the probability to select the correct response. Consequently, a generally enhanced and earlier N2 / N450 indicates better performance.
In the above cited cross-sectional study by Gajewski et al. (2010a) with younger and older employees, older assembly-line workers had a lower switch performance than flexibly working older employees (i.e., without a fixed work routine) who showed a similar switch performance to the young workers. In a follow-up, randomized, controlled training study Gajewski et al., 2017a, older assembly line workers from the same factory completed three months of teacher-guided cognitive training two times a week. A waiting control group was included that received the same training after the first group finished its training. After the training, the trainees showed clear improvements in performance in the switch task, particularly reduced error rates in the memory-based mixing block. The ERPs showed an amplitude increase in the N2 that remained stable three months after the training was finished. The same N2 increase was observed in the waiting control group after training (Fig. 4 top) which validated this finding. Similar N2 increases were observed after four months of twice weekly cognitive training in seniors who were 70 years old on average Küper et al., 2017). Fig. 4 shows the training-related increase in the N2 amplitude in middle-aged (top) and older individuals (bottom).
These results were recently supported by Gaál and Czigler (2017) and Olfers and Band (2018). Gaál and Czigler (2017) administered eight hours of cognitive training to younger and older adults. Both young and old training and control groups were evaluated one month later. Directly after the training, the mixing costs of the older trainees were lowered, reaching a similar level as observed in young adults. Moreover, this mixing cost reduction was maintained one year later. The training-related gains were accompanied by an increase in the N2 amplitude after training that remained larger compared to the pre-test in a one year follow-up study, while no changes were found in the control groups. Olfers and Band (2018) administered video-game training in young participants for 6 weeks and found benefits in task performance and an enhanced N2 after training relative to controls, corroborating the findings in older and middle-aged adults.
Finally, Gajewski and Falkenstein (2015a) recently reported an association between life-long physical activity and task-switching performance. Two matched groups of older men about 70 years participated in the study. The participants differed significantly in respect to their habitual long-term physical activity. The physically active participants showed lower mixing costs, whereas no differences in switch costs were obtained. This difference was accompanied by a larger N2 in the active seniors compared to the sedentary group. A corresponding effect was observed on the N450 amplitude in a Stroop task (Gajewski and Falkenstein, 2015b).
In sum, these results indicate that performance during task switching in older age is affected by different endogenous or exogenous factors. The performance differences are mainly apparent in mixing costs in speed, accuracy, and occasionally in variability of speed, suggesting inter-individual differences in the ability to maintain and to apply multiple task sets. Superior performance is accompanied by increased amplitude of the N2 or the N450, and/or a reduced N2 latency. Cognitive training improves task-switch performance and increases the N2 presumably due to reduction of task interference. Thus, the N2 / N450 seems to play a pivotal role in the implementation of task sets and may serve as an index of neuronal integrity in older age.

Allocation of cognitive resources (Target-P3)
By far, the most frequently investigated ERP component in the target phase during task switching is the P3. The target-P3 is a large positive wave that follows the N2 with a peak between 400 and 600 ms post-target. Similar to the cue-locked P3, the target-locked P3 in taskswitching studies has a parietal focus in young adults and a frontoparietal distribution in older age (Fig. 3) (Adrover-Roig and Barceló, 2010;Gajewski et al., 2010a;Friedman et al., 2008;Karayanidis et al., 2011;Kopp et al., 2014). However, cue-and target-P3 have different functional properties .
It has been demonstrated that the P3 in reaction tasks occurs with similar or even larger amplitude in response-locked averages (Berchicci et al., 2016;Friedman et al., 1978;Friedman, 1984;Gajewski et al., 2016a, b;Verleger et al., 2005Verleger et al., , 2014Whitson et al., 2014). Thus, the target-P3 has been related to integrative processes between stimulus processing and response execution and may be associated with post-decisional processes (Nieuwenhuis et al., 2005;Verleger et al., 2005), like an after-effect of response selection and/or activation. The reduction of the target-P3 in task switch relative to non-switch conditions is not due to a latency jitter of the target-P3b but reflects a genuine amplitude reduction because it is observed both in stimulus-as well as in response-locked ERPs Falkenstein, 2011, but see Whitson et al., 2014). 2 Karayanidis et al. (2011) found that mixing and switching effects on the earlier part of the target-P3 (400-600 ms) were especially pronounced and widespread in an older group between 60 and 79 years. The mixing effects increased with age, whereas the differentiation between non-switch and switch trials seems to be age-invariant. In a subsequent study using the same data set, Whitson et al. (2014) aimed at investigating the locus of residual and mixing costs as a function of age by analyzing target-and response-locked P3. They observed large mixing effects both on the target-and response-locked P3 in the oldest group, whereas a local switch effect was obtained in the target-locked P3 only. The authors concluded that in older adults, the increase of mixing costs was associated with greater sustained interference at the level of decision-response mapping for non-switch trials, whereas residual switch costs were due to stimulus-level interference. Adrover-Roig and Barceló (2010) investigated middle-aged and old adults subdivided in low and high cognitive control participants. They reported reduced P3 amplitudes on switch trials compared to deviant distractors across all groups except for the low control older group, suggesting a multiplicative interaction between age and cognitive control for the neural substrates of switch costs. A similar approach was chosen by Goffaux et al. (2008) whereby authors investigated the effect of aging and working-memory capacity on task-set switching, by dividing the sample in groups of high and low working-memory performers. Old low performers showed significantly larger mixing costs in speed than young and old high performers. However, analysis of the target-locked P3 did not reveal any group effects or interactions with trial type. Instead, as described above group differences were observed in the preparation interval, suggesting the use of proactive task control in high performers prior the task implementation. De Sanctis et al. (2009) also compared young participants with low and high older performers subdivided by a composite measure of performance from different executive tasks. Older adults with high levels of executive functioning showed a delay of the target-P3, but with a similar pattern as in young adults, i.e., the P3 was enhanced on pre-switch trials and reduced on switch trials, as compared to no-switch trials. In contrast, older participants with low levels of executive functioning showed no evidence of this differentiation across trial types. In this vein, Gaál and Czigler (2015) found an enhanced target-P3 on non-switch than switch trials in younger adults. This differentiation as well as the absolute amplitude of the P3b was strongly reduced in older adults. A recent study by Enriquez-Geppert and  analyzed target-P3 and time-frequency decomposition of EEG signals and used a cueing paradigm in young, middle-aged, and old participants. They found P3 amplitude and power reductions as a function of age extracted from topographically and functionally different networks. The most extensive age-related 2 These different results may be due to different tasks applied in both studies. Whereas  used exclusively bivalent stimuli, providing a full overlap of stimulus-response sets, i.e., in each task the same digit was categorized either in respect to its number or to its parity, Whitson et al. (2014) used two univalent stimuli simultaneously presented (letter and digit) that required categorization of either the one or the other stimulus, which led to a lower overlap, and therefore may have led to lower interference during response selection; cf. Gajewski et al., 2010b). changes were found in a parietal network associated with mixing and restart processes.
As outlined above, Schapkin et al. (2014) compared middle-aged and younger adults using cue-based (CTI = 0 ms) and memory-based task switching and found larger mixing costs in older than younger adults that did not vary as a function of working-memory load. Unexpectedly, larger switch costs were obtained in the younger group in the memory-based block. The corresponding target-locked P3 showed the typical reduction of amplitude on switch vs. non-switch trials in both groups and confirmed the general age-related reduction and delay of the target-P3. Interestingly, the target-P3 latency was significantly delayed in the older but not in the younger group for non-switch relative to switch trials. These results indicate that increased mixing costs in the elderly are not completely explained by deficits in advanced taskset updating, working-memory load, and preparation efficiency, but are also partly due to a delay in response selection and execution due to task-set interference in mixing blocks (Karayanidis et al., 2011;Mayr, 2001;Whitson et al., 2014).
In the study by Friedman et al. (2008), age effects were found in switch costs in speed but no effect was observed regarding mixing costs. They reported an interaction between trial type and age group on the target-P3 amplitude, suggesting that in the younger age group switch trials elicited the largest amplitude compared to all other trial types. By contrast, for older adults, switch-trial amplitude did not differ from non-switch amplitude of the target-P3. This was interpreted as an evidence for a successful update of the new task set in younger participants, whereas in the older group the updating of task sets may have occurred on both switch and no-switch trials. In the above mentioned study by Gajewski et al. (2010a) with younger and older employees with and without repetitive type of work, older workers with repetitive work showed larger mixing costs than flexibly working older employees in memory-based but not cue-based task switching. No age or job-related effects on the local switch costs were found. This pattern was accompanied by decreased target-P3 in the mixed block compared to the single-task block and was more strongly reduced in older assembly line workers than younger ones. Regarding local effects, target-P3 was reduced on switch trials compared to non-switch trials in the memorybased but not in the cue-based block. The absolute target-P3 amplitude was attenuated in the older assembly line group compared to all other groups, suggesting that lower performance was associated with reduced target-P3, and that this reduction and performance decrements occur when working-memory demands are high (Gajewski et al., 2010a(Gajewski et al., , 2016bHohnsbein et al., 1998). In addition, employees from stressful occupations with subclinical burnout (i.e., with burnout symptoms but without a clinical diagnosis) show an inverse relationship between burnout severity, performance, and target-P3 amplitude in memorybased task switching (Gajewski et al., 2017b). The inverse relationship between performance and P3 amplitude is in line with a training-induced increase of the target-P3 observed in the described above training studies in seniors and middle-aged workers (Gaál and Czigler, 2017;Gajewski et al., 2017a). This is also in line with findings obtained by Hillman et al. (2006) who analyzed performance of younger and older physically active and low-active participants using a cued task-switching paradigm without the opportunity to prepare the task (CTI = 0 ms). They found smaller mixing and switch costs as well as shorter latencies and larger amplitudes of the target-P3 in the more physically active older adults (see also Tsai and Wang, 2015 for similar findings).
However, a reverse effect was observed in the study by Gajewski and Falkenstein (2015a) reporting enhanced performance in a physically highly active group of seniors that was accompanied by a more negative N2 and lower target-P3 than in sedentary seniors. The superior performance in this task was supported by a clear differentiation between switch and non-switch trials in the physically active seniors, whereas the target-P3 amplitude did not differentiate between switch and non-switch trials in the low active group. An increased N2 and reduced subsequent target-P3 was also found after cognitive training in a recent study by Küper et al. (2017). Interestingly, N2 and target-P3 amplitude were highly correlated in former studies , indicating that the more negative the N2, the smaller the target-P3. This suggests that both components are not fully independent of each other and may be superimposed on a trailing edge of a long lasting sustained negative wave.
Taken together, several pieces of evidence lead to the conclusion that enhanced performance in the switching task is related to larger target-P3 amplitude. However, this observation is not fully consistent across the literature. The discrepancies may be partly due to task demands or different switch paradigms, for example, memory-based vs. cue-based task switching or using long vs. short intervals for advanced preparation. Particularly, the temporal and contextual predictability of targets seems to play an important role as uncertainty modulates the amplitude of the target-P3 (Barceló & Cooper, 2018;Eppinger et al., 2017). Therefore, the P3 amplitude difference between switch and nonswitch trials seems to be more indicative for high performance than the absolute amplitude of the target-P3. In most of the studies cited, the size of this switch effect was related to performance and was typically observed in young (Barceló et al., 2000(Barceló et al., , 2002Kieffaber and Hetrick, 2005;Karayanidis et al., 2003Karayanidis et al., , 2011Lorist et al., 2000;Nicholson et al., 2005;Poulsen et al., 2005;Rushworth et al., 2002Rushworth et al., , 2005 and older high performers (Adrover-Roig and Barceló, 2010;De Sanctis et al., 2009;Gajewski and Falkenstein, 2015a;Tsai and Wang, 2015) and was less evident or absent in low performers.
In sum, target-locked P3 associated with a closure of a cognitive epoch and allocation of cognitive resources to the task is reduced in older age. This is accompanied by lower performance during task switching, mainly in mixing blocks that include interference. The frontal shift of the target-P3 with age may suggest recruitment of frontal resources to manage the task, or alternatively changes in neuronal networks and brain morphology in aging producing similar P3 patterns at frontal and parietal areas. Indeed, increased frontal positivity may be interpreted as reduced negativity following loss of brain tissue in frontal areas.

Summary: ERP during task implementation
The target-P2 seems to be the first target-locked component differentiating between repeat and switch trials, but aging studies are scarce. The reported relationship between the P2 and mixing or switching costs suggests that the target-P2 is involved in retrieval of the currently relevant stimulus-response mappings that have to be implemented to perform the task. The P2 appears to be enhanced in older vs. young subjects in mixing blocks whereas no difference occurs in single task blocks. Both P2 amplitude and latency can be reduced by training.
The subsequent N2 is usually larger and/or delayed in switch than non-switch trials. Moreover, the amplitude of the N2 decreases with age. The N2 differentiation between switch and non-switch trials has been reported to be either similar across all age groups or even larger in older subjects. This observation accords with the age-invariant switch costs in behavior. The N450 (and Ni) with similar functional properties as the N2 was more pronounced in incompatible than compatible Stroop trials and showed lower amplitudes and delayed latencies in older than younger adults, suggesting impaired interference processing. Some task-switching studies in elderly adults aimed at investigating the functional properties of the N2 in more detail as a function of exogenous (environmental) and endogenous (biological) factors. The results showed that larger amplitudes of frontocentral negative components like N2 and N450 were associated with enhanced performance.
Generally, the target-N2 plays a crucial role during reactive control of task sets. The rather consistent correlation with performance suggests that it reflects a real-time index of overt performance. The target-N2 seems to represent a response-selection mechanism that is modulated by task demands. Increased task demands like trials in mixed-task blocks seem to increase the threshold for selecting the correct response and reduce the averaged N2 due to increased variability in the selection process. Aging and other unfavorable factors may elevate this threshold, resulting in slower responses, lower accuracy, and increased intra-individual variability of performance. This is presumably due to larger neuronal noise and lower integrity of the neuronal system. This deficit can be partly diminished by cognitive or physical training.
Finally, the target-locked P3 during task switching has been consistently found to be reduced and delayed with increasing age. Additionally, the P3 topography in aging is more equally distributed across parietal and frontal brain areas. Moreover, target-P3 is smaller in mixed than single task blocks and smaller in switch than non-switch trials, consistent with the observation that increasing task demands reduce the target-P3 amplitude. This specific differentiation was mainly found in younger adults and older high performers, whereas older low performers show the similar amplitudes in non-switch and switch trials, suggesting impaired discriminability between the trial types. Finally, performance deficits during task switching as well as the target-P3 attenuation in older age can be partly ameliorated by cognitive training. This is in line with evidence, suggesting an association between higher performance and larger target-P3 amplitude.
In sum, older persons show a delayed and less efficient retrieval of task instructions (P2), a compromised implementation of task sets, resolution of interference and response selection (as indicated by a smaller, and delayed N2 or N450) and reduced allocation of cognitive resources to the task (indexed by the target-P3) compared to young adults.

Correct response activity (Nc / CRN / MFN)
A small number of task-switching studies analyzed ERPs timelocked to the overt response. Two types of trial events can be distinguished in these response-locked analyses: ERPs in trials with correct and in trials with erroneous responses (Fig. 5). In correct trials, there is a negative deflection with a frontocentral distribution peaking between 70 and 130 ms after response onset, termed correct negativity (Nc; Falkenstein et al., 2000;Yordanova et al., 2004a) or correct-response negativity (CRN; Coles et al., 2001;Gehring and Knight, 2000). Current views on this component suggest that it reflects response conflict monitoring and detection Gehring and Fencsik, 2001;Kray et al., 2005;Ridderinkhof et al., 2004).
In the above reported study, Kray et al. (2005) analyzed age effects of the Nc using compatible and incompatible Stroop trials and found no group differences in Nc latency between younger and older adults. However, there was an amplitude increase in incompatible as compared to compatible trials. This effect was highly significant in the younger age group, whereas there was no effect of compatibility in the older age group. This was interpreted in terms of an age-related decline of the action monitoring system, i.e., while younger adults seem to detect more conflict on incompatible trials, older adults appear to process both trial types in the same manner (see also West and Moore, 2005, for similar results). Eppinger et al. (2007) also investigated the Nc using the same task-switching paradigm with low frequent incompatible Stroop stimuli. They replicated the findings of their previous study (Kray et al., 2005) and extended them by showing an inverse relationship between conflict frequency and Nc amplitude at least in younger adults, suggesting that older adults are less able to adjust their behavior in conflicting situations. Friedman and coauthors (2008) reported results on a similar negativity that they termed mediofrontal negativity (MFN, cf. Gehring and Willoughby, 2002) as a function of age and task-switching demands. In the young group, the MFN amplitude increased gradually from no-switch to pre-switch to switch trials, whereas this effect was not apparent in older adults. This confirms the results by Kray et al. (2005) and Eppinger et al. (2007). Interestingly, Friedman et al. (2008) observed that the MFN was preceded by a positivity occurring prior to the response that was inversely related to the MFN, i.e., with smaller amplitudes in more demanding conditions like switch trials compared to non-switch and pre-switch trials. This positive wave apparently reflects the response-P3, which was equivalent to the target-P3 that was found to be smaller on switch than non-switch trials .
In sum, the Nc / CRN is a negative response-locked potential sensitive to response demands or response conflict. Its amplitude is larger on switch than on non-switch trials in younger participants, while this difference disappeared in the elderly. A similar pattern was observed regarding the MFN, which is more pronounced on switch than nonswitch trials in young participants only. These findings suggest decreased response monitoring during switching trials in older age.

Error response activity (Ne / ERN)
The second type of response-locked studies focuses on ERPs timelocked to an incorrect response (Fig. 1). Similar to the topography, timing, and polarity of the Nc / CRN / MFN, a prominent negative wave occurs on error trials, which is generally larger on error than correct trials. This component has been termed error negativity (Ne; Falkenstein et al., 1990Falkenstein et al., , 1991 or error-related negativity (ERN; Gehring et al., 1993) and reflects detection of errors or conflict of response tendencies on a level of monitoring motor responses (Falkenstein et al., 2000;Yeung et al., 2004). The amplitude of the Ne is generally reduced in older age (Falkenstein et al., 2001;Hämmerer et al., 2014). Following the Ne, a large positive wave is visible (Pe) that is associated with evaluation of the erroneous response and is also clearly attenuated in older versus younger adults (Falkenstein et al., 2001). Fig. 5 presents the Ne and Pe in younger and older individuals in the task-switching paradigm. West (2004) also analyzed the Ne in younger and older adults separately for correct incompatible and intrusion errors in Stroop trials (reading the word instead of naming the color). As found in other paradigms, the Ne amplitude was reduced in older compared to Fig. 5. An example of response-locked ERP collapsed across task repetition and switch trials for correct (black) and erroneous responses (red) in young (left), and old individuals (right). The relevant components are labelled. Negativity is plotted upward. Unpublished data (For interpretation of the references to color int his figure legend, the reader is referred to the web version of this article). younger individuals. Additionally, the differentiation between error trials (Ne) and correct trials (Nc) was absent in older participants (cf. Fig. 5). This shows alterations of error processing in aging (Falkenstein et al., 2001).
Only a few studies using a task-switching paradigm analyzed the Ne, and aging studies are particularly scarce. Moreover, ERP were often collapsed across switch and non-switch trials to reach a reasonable number of erroneous trials. Themanson et al. (2006) analyzed the Ne in a task-switching paradigm in older and younger adults in correspondence with their physical activity. In addition to larger mixing and switching costs in elderly than in younger participants and lower mixing costs in physically active than less active older adults, they found that the amplitude of the Ne was smaller in older as compared to younger participants. Surprisingly, the amplitude of the Ne was associated with mixing costs, i.e., an increased Ne was associated with increased mixing costs. This relationship was observed only in the highly active but not less active participants. Similarly, in the above-mentioned study by Gajewski et al. (2010a), the Ne was analyzed as a function of age and type of job. The Ne was clearly attenuated in older assembly line participants (showing lowest task-switch performance) compared to the other groups in the memory-based condition, while this reduction was only marginal in the cued condition. This observation is inconsistent with the results obtained by Themanson et al. (2006) who demonstrated reduced Ne amplitude in physically active participants who showed higher performance. However, in the Themanson study, the entire response-locked ERP appears to be shifted to more positive values in active participants, suggesting that the effect is not specific to the Ne. Also, in the above cited training study in seniors,  found reliable improvement of performance during task switching and an increase of the Ne amplitude in post-relative to pre-test in the cognitive training group. This effect did not occur in other groups (physical, relaxation, no-contact). The same training effect on Ne was observed in another sample of middle-aged assembly-line workers who were trained using similar cognitive training. The Ne increase remained stable until 3 months after the training was finished (Gajewski et al., 2017a). These results suggest that a larger Ne is related to better performance in task switching at least in older subjects.
Taken together, the Ne / ERN was analyzed in few task-switching studies. The Ne showed reduced amplitude and diminished differentiation between the amplitudes on error vs. correct trials in elderly participants. Yet, the results are not fully consistent: whereas one study found that higher performance in physically fit subjects was associated with lower amplitude of the Ne, some other studies have shown larger Ne amplitudes in subjects with high task-switch performance. Switching performance can be improved due to cognitive training, which is related to an enhancement of the Ne. This suggests that the Ne observed in switch tasks is related to behavioral switching costs, i.e., the larger the Ne, the better the switching performance. Therefore, it is plausible to assume that the increase in the Ne is a consequence of an efficient response selection, as reflected in the N2. In other words an improved implementation of stimulus-response sets (for example by training) induces a higher awareness about a required response which results in a larger Ne in erroneous responses.

Summary: ERPs during response monitoring
In sum, a number of recent task-switching studies investigated age effects on the response-locked negativities related to conflict processing (Nc / CRN) and error detection (Ne / ERN). In general, the amplitude of the Ne is reduced in older age. Conflicting stimuli evoked larger Nc and Ne amplitudes than compatible ones, but this effect is diminished in older adults, suggesting that elderly are less able to detect errors and adjust the behavior according to the task context.
3. Implications of the ERP findings and integration into models of executive control: how can ERP research contribute to define these models?
The studies reported above show that one of the major contributions of ERPs has been that they make it possible to separately examine preparatory, task implementation, and response processes. This led to new insights into how cognitive control and action selection work. In the preparatory interval, older subjects show a less efficient or inadequate updating on switch trials, as reflected in a reduced cue-P3 difference between switch and non-switch trials and a general attenuation of cue-P3 in mixing blocks, whereas the cue-P3 in the single task blocks does not show age-related changes. This suggests that preparatory processes in a complex task context are altered in the elderly (e.g., Kray et al., 2005;West and Moore, 2005;Karayanidis et al., 2011;Kopp et al., 2014). This deficit is most probably compensated for by a stronger recruitment of frontal resources and a stronger maintenance of tasks-sets in switch trials. This explains the absence of age differences in local switch costs as a function of preparation time (Cepeda et al., 2001;Hahn et al., 2004;Hartley et al., 1990;Kramer et al., 1999;Kray, 2006;Mayr and Liebscher, 2001;Meiran et al., 2001). The ERP effects after the target stimulus suggest that the performance decrements in older age are partially due to altered S-R retrieval from working memory (as reflected in the P2), but mainly due to deficits in interference processing during response selection, as reflected in the N2 and related negative components (N450 and Ni; Eppinger et al., 2007;Kray et al., 2005;West, 2004). The N1 and P2 are increased, while the N2 / N450 decreased and often delayed in older vs. younger subjects. This deficit can be temporarily compensated via an increased use of resources and additional effort at the time of task implementation and response monitoring (De Sanctis et al., 2009;Karayanidis et al., 2011) or restored by physical or cognitive training (Gaál and Czigler, 2015;2015a, 2015bGajewski et al., 2017a;Küper et al., 2017;Olfers and Band, 2018;Wild-Wall et al., 2012). In particular, the P2 amplitude and latency can be reduced, and the N2 and Ne / ERN amplitude enhanced due to cognitive training in middle-aged and older adults, which is accompanied by a substantial improvement in behavior.
Many theories of cognitive aging postulate that a decline in cognitive control leads to age-related impairments in performance. Braver and Barch (2002) assume that flexible behavioral adaptation is impaired in old age because the ability to maintain context information in working memory as well as its updating and the shielding against interference is impaired. This view is supported by changes in the frontocentrally distributed ERPs associated with working memory and interference processing. Moreover, the findings demonstrate that the PFC and the dopaminergic system are affected by a strong age-related decline (Bäckman and Farde, 2005;Bäckman et al., 2000Bäckman et al., , 2006Bäckman et al., , 2010Raz, 2000;West, 1996). These observations are also in line with studies finding a performance decrement in the elderly mainly in tasks that rely on the PFC and the dopaminergic system. This involves not only task switching per se (Nagano-Saito et al., 2008; for a meta-analysis, see Wasylyshyn et al., 2011; for a review, see Kray and Ferdinand, 2014) but also working-memory (Hale et al., 2007;Reuter-Lorenz and Sylvester, 2005) and action monitoring mechanisms (Eppinger and Kray, 2011;Ferdinand and Kray, 2013;Hämmerer and Eppinger, 2012;Nieuwenhuis et al., 2002) that are crucially involved in activation, preparation, and implementation of task sets.
This view of different components contributing to task performance has also been put forward in the dual mechanisms of control (DMC) theory by Braver and colleagues (Braver et al., 2007). They differentiate between a proactive and reactive control mode. Proactive control denotes a form of early selection in which goal relevant information is actively maintained to ensure optimal task preparation. In contrast, reactive control is characterized by late correction processes that are applied whenever necessary. This is the case, for instance, when an error has been committed or conflict is detected during responding. Thus, a proactive control mode is associated with enhanced advance preparation, while a reactive control mode is related to the resolution of interference as soon as it is detected (see also Jacoby et al., 1999). The preferred control mode may vary intra-individually, but also depending on the task or situation. It is also dependent on PFC maturation and thus on the development of cognitive control. Braver and colleagues assume that this is the reason why older adults use a more reactive than proactive control mode in comparison with younger ones. These results are corroborated by ERP findings demonstrating that older adults exhibit less efficient and less flexible preparatory processes after cue presentation and more conflict processing during responding in the AX-CPT (Schmitt et al., 2014a, b). Additionally, fMRI studies showed that these different control modes are associated with temporal shifts in PFC activation patterns. Younger adults' proactive mode is related to a stronger activation in lateral PFC after cue presentation, while older adults' reactive mode is linked to heightened activation of the lateral PFC after the probe is presented (Jimura and Braver, 2010;Paxton et al., 2008). However, it is important to note that the time scales between fMRI and ERP are different and the activation patterns are not directly comparable.
Also, as shown above, ERP studies using the task-switching paradigm differentiate temporarily between proactive and reactive control and would be especially suited to help shed light on these age-related temporal shifts in advance preparation vs. task implementation and action monitoring. However, to date only few ERP task-switching studies exist that explicitly test how proactive and reactive control processes interact and change in old age. One of these studies was conducted by Kopp et al. (2014) who investigated age-related trade-offs between proactive and reactive strategies of cognitive control. They observed a deficit in proactive control in elderly individuals at the expense of higher loads on reactive control in the cue-locked ERPs, indicating that in contrast to older adults, younger ones prepared the task in advance. Nevertheless, local switch costs in performance did not vary between the groups and the targetlocked P3 did not differ as a function of age groups and switch demands. In a similar vein, Karayanidis et al. (2011) found age-related increases in the intensity of ERP indices of preparation for mixed-repeat trials, anticipatory attention (frontal shift), and post-target interference, consistent with increased, but less efficient proactive and increased reactive control processes in older adults.
The present overview on behavioral and ERP data shows that both proactive and reactive processes in task switching are altered in older subjects. As to proactive control, task-set management appears to be inefficient (i.e., similar updating of task-sets on switch and non-switch trials). As to reactive control, task-set retrieval and response selection in mixing blocks are clearly impaired in older subjects, presumably due to increased susceptibility to interference. Moreover, response monitoring and error detection are attenuated. Both results clearly show deficits in reactive processing in mixing blocks where the cognitive demands due to interference are high.
The mechanisms underlying processing of targets and translation into overt behavior seem to be more complex and thus more vulnerable to cognitive aging than processes involved in voluntary proactive control. This assumption was also supported by the findings that reactive processes were mainly enhanced by different training regimes, which also improved performance. In contrast, to the best of our knowledge, there is to date no evidence in the task-switching literature that enhancement of proactive mechanisms can substantially improve performance in older adults. Thus, deficits in reactive control can be ameliorated by regular physical and cognitive activity and training that illustrates plasticity of executive functions in older age.

Limitations and future prospects
Firstly, the task-switching paradigm and the above reported results also demonstrate that proactive and reactive control are interdependent processes and that it is important to further examine their interplay (Cooper et al., 2015). In the recent literature, however, those two aspects are treated rather independently in different research areas and models. Although models of cognitive control usually include a socalled gating mechanism that initiates the updating of working-memory contents if relevant information is detected, and that acts as a trigger for the implementation of cognitive control, there is usually no explicit link to models of performance monitoring. Similarly, in models of error or conflict processing, it is only vaguely described how the detection of conflict or a prediction error can actually lead to adjustments of cognitive control and therefore changes in behavior. Thus, on a conceptual note, the task-switching paradigm and the above reported results demonstrate that there is a need for theoretical models that can integrate cognitive control and performance monitoring and account for age-related changes in these processes.
Secondly, it is important to emphasize the fact that there is obviously no single mechanism or simple one-to-one correspondence between specific behavioral and electrophysiological signatures associated with cognitive aging. Thus, complex and dynamic interactions between the observed behavioral deficits and different stages of neural processing as reflected by functionally distinct electrophysiological indexes would be a more proper way to understand the source of the agerelated decline in executive functions. Also, further biological (genetic, inflammatory, infectious etc.) and environmental (educational, cognitive, physical, social etc.) factors contributing to the age-related effects in executive functions should be considered in the models. Thus, adopting multivariate techniques and multidimensional theoretical approaches in future studies would be necessary to enhance the understanding of cognitive aging (Park and Festini, 2017). Additionally, it would be essential to consider intra-individual variability of behavior and neural processing in the model of executive functions in aging as they provide important information about the quality of cognitive processing and reliable prediction of a risk of cognitive impairment or dementia (MacDonald et al., 2006).
Thirdly, an important issue is related to the temporal overlap of ERP components. Using conventional techniques, the temporally and functionally distinct processes cannot be completely disentangled. The electrophysiological responses during target and response processing overlap to a large degree and should be analyzed in the future by more recent methods, such as Independent Component Analysis (ICA) or Residue Iteration Decomposition (RIDE), allowing trial-by-trial decomposition of the EEG signals (Brydges and Barceló, 2018;Ouyang et al., 2015;Verleger et al., 2014Verleger et al., , 2017. Finally, further studies are necessary 1) to replicate training-induced gains in performance, 2) to document stability of training effects and "rejuvenation" of older brains, and 3) to investigate changes in executive functions across the entire lifespan. Simple comparisons between younger and older adults are methodically insufficient to allow causal relationships to be discovered. Therefore, it is crucial for our understanding of executive control across the lifespan to include and analyze additional environmental and biological factors and their interaction in a longitudinal study design. This would substantially help to better understand the dynamics of cognitive aging.

Conclusions
The task-switching paradigm is a promising tool to investigate agerelated changes in executive functions. It allows analysis of different aspects of information processing and helps to identify the most vulnerable cognitive processes that are affected by aging. Previous research showed age differences during task preparation, task implementation, and action monitoring. ERPs play a crucial role in detecting different processing steps influenced by aging due to an excellent temporal resolution, a reasonable spatial resolution, and the possibility to use the EEG signals to apply a variety of methodological approaches and to conduct sophisticated data analyses. Thus, ERP research combined with the task-switching paradigm has obtained increasing interest in the domain of cognitive neuroscience and aging research in the last decade and offers further potentialities for development.

Conflict of interest
The authors have no conflicting interests to declare.