The effects of age-bias on neural correlates of successful and unsuccessful response inhibition

Response inhibition is important for adherence to social norms, especially when norms conflict with biases based on one ’ s social identity. While previous studies have shown that in-group bias generally modulates neural activity related to stimulus appraisal, it is unclear whether and how an in-group bias based on age affects neural information processing during response inhibition. To assess this potential influence, young adults completed a Go/NoGo task incorporating younger face (in-group) and older face (out-group) stimuli while undergoing functional magnetic resonance imaging (fMRI). Our results replicated previous findings by demonstrating higher accuracy in successful Go compared to NoGo trials, as well as the engagement of nodes of the response inhibition network during successful response inhibition, and brain regions comprising the salience network during unsuccessful response inhibition. Importantly, despite a lack of behavioural differences, our results showed that younger and older face stimuli modulated activity in the response inhibition and salience networks during successful and unsuccessful inhibition, respectively. Interestingly, these effects were not uniform across net- works. During successful response inhibition, in-group stimuli increased activity in medial prefrontal cortex and temporo-parietal junction, whereas out-group stimuli more strongly engaged pre-supplemental motor area. During unsuccessful response inhibition, in-group stimuli increased activity in posterior insula, whereas out-group stimuli more strongly engaged angular gyrus and intraparietal sulcus. Consequently, the results infer the presence of an age-bias effect in the context of inhibitory control, which has substantial implications for future experimental design and may also provide the means of investigating the neural correlates of implicit beliefs that contribute to ageism.


Introduction
Response inhibition is a crucial element of executive control, encompassing the suppression of information and related actions to support goal-directed behaviour in dynamic environments [88]. A range of deficits are associated with failure to inhibit a response, which can result in loss of sustained attention, increased distractibility, or impulsive behaviour [8,38,40]. In a social context, response inhibition forms part of an individual's ability to conform to social norms; in particular, norms that conflict with goals, beliefs, or attitudes, which are based on the individual's social identity and manifest as in-group bias.
Failure to inhibit in-group bias can perpetuate existing stereotypes, such as ageism, i.e., explicit actions or implicit attitudes that discriminate against older adults [9]. Analysis of written accounts by younger adults engaged in a senior mentoring programme revealed instances of negative discriminatory language towards older adults, while perceived attributes of younger adults were viewed in a favourable manner [27]. Failure to constrain such in-group versus out-group beliefs has been shown to be damaging to older adults themselves and to hinder active ageing due to the internalisation of these negative views [82]. This phenomenon also extends to a professional setting, where attitudes towards older adults held by health care workers have been established to produce discrimination in rehabilitation services [69]. It is, therefore, crucial to understand the origins of age-based in-group bias and particularly its interaction with response inhibition, in order to prevent its detrimental consequences. Neuroscientific studies have demonstrated that in-group bias generally modulates brain activity and is associated with differences in brain systems that are specific to the stimulus or task, such as the amygdala and fusiform gyrus in response to face stimuli [29,94], medial prefrontal and anterior cingulate cortices during conflict resolution [19,25], the salience network during empathy [97], or the temporo-parietal junction during mentalising [86]. However, to date, no study has addressed the question of whether in-group bias modulates brain activity in regions associated with response inhibition.
Studies examining response inhibition most commonly utilise the Go/NoGo task [22]. Considered to be a measure of action restraint (as opposed to action cancelling, for which the Stop-Signal task is advocated: [72,81]), Go/NoGo requires participants to respond to standard or frequent events (Go trials), but withhold their response to novel or infrequent events (NoGo trials). A network comprising fronto-parietal structures has consistently been reported to underlie such action suppression processes, with right lateral frontal cortex thought to play a critical role [57]. NoGo stimuli have been shown to activate right inferior frontal gyrus (IFG; [4]) in conjunction with the subthalamic nucleus [5], supplementary and pre-supplementary motor areas (SMA; [75]), premotor cortex [91], and subregions of the parietal cortex, such as inferior parietal lobule (IPL; [67]). Dorsolateral prefrontal cortex (dlPFC) activity is also often observed, although this may better reflect the influence of cognitive processes that aid inhibition, such as conflict monitoring [30], attention [32], working memory [57], and response selection [75]. Error detection, for example, is a distinct but related construct; resulting in largely overlapping activation while also triggering additional recruitment of right anterior cingulate and insula cortices [55,74]. Consequently, the pattern of neural co-activation, corresponding to the implemented behavioural paradigms, can be dramatically different depending on the nature of the stimuli; particularly where complex, top-down intensive tasks are utilised (often involving multiple Go and NoGo cues, [15,54,89]).
Go/NoGo tasks have incorporated a variety of stimuli; including numbers, letters, and images ( [20,44,57]. Within the image-based investigations, faces are particularly interesting for studies focusing on in-group bias. Facial perception is considered one of the most developed visual skills possessed by humans [35], with faces representing a unique yet commonly viewed stimulus that can express a multitude of information (spanning identity, emotion, age, and gender; [66]). Therefore, faces provide vital cues required to successfully navigate social interactions [48]. Individuals often form static impressions of others based on superficial characteristics, such as appearance. For example, visual cues related to age have consequences for impression formation that can perpetuate negative stereotypes and ageism, in relation to judgments of mental and physical capacity. Studies of such 'diagnostic facial cues' [100] have shown that younger adults have an innate bias against older adults founded on facial appearance [43]. This implicit value judgement corresponds to research that supports the assertion that expertise in face perception primarily applies to faces that are particularly salient [98], as opposed to humans being equally adept at processing all faces [11].
Accordingly, there appears to be a discernible processing advantage for same age faces [46], indicating that individuals tend to attribute greater importance to faces of those from their own demographic. Young participants, therefore, find it easier to recognise and distinguish between young faces [2], compared to faces from other age groups [37]. Neuroimaging studies suggest that greater activity in medial prefrontal cortex (mPFC), insula, and amygdala for same-age, compared to other-age faces, underlies this face processing preference [23]. Involvement of mPFC has also been demonstrated to be independent of emotional expression, indicating its role as fundamental to the own-age bias [102]; a finding which is highly consistent with what is known of the mPFC in relation to salience within the bounds of social interactions [7,79]. Taken together, these results suggest that the observed bias in face processing may provide a unique opportunity to study inhibitory control in the context of interactions with other-age individuals.
In the context of the present study, a Go/NoGo paradigm with facial stimuli of younger and older adults was used to investigate the effects of age bias on response inhibition. Specifically, we aimed to determine whether the presentation of in-group stimuli (younger faces) and outgroup stimuli (older faces) modulated neural activity during successful and unsuccessful response inhibition in a sample of young adults. Participants were asked to respond to the present stimulus unless it matched the previous one (separated by a brief interval of 500 ms); making the task far less complex than those, for example, where participants must choose between two possible NoGo cues after intervals of several seconds [26]. Therefore, we considered working memory demand to be low, such that we were able to produce a simple manipulation of stimulus salience and minimise additional cognitive strategies, to better explore response inhibition in as isolated state as possible [77]. To our knowledge, there have been no such attempts to create a Go/NoGo task in this manner, yet this approach may provide the insight necessary to determine activity in subregions of the response inhibition network as being contingent upon specific stimulus attributes. In accordance with the literature, we expected (1) higher accuracy on Go trials than NoGo trials and (2) superior behavioural performance, particularly with regard to reaction time for trials featuring younger, compared to older, adult faces. We further anticipated (3) to replicate previous findings showing increased activity during correct NoGo trials in regions essential for response inhibition (e.g., parietal cortex, IFG, and pre-SMA), and during incorrect NoGo trials in structures associated with error detection (e.g., anterior cingulate cortex and insula). Finally, we predicted (4) that own-age face bias would modulate activity in mPFC, by establishing stronger responses to younger compared to older adult face stimuli.

Participants
We tested 46 right-handed young adults with normal or corrected-tonormal vision. The final sample size was 44 (mean age = 24.27 years; SD = 3.55 years, 23 males) as two participants were excluded from all analyses due to a lack of understanding of the task instructions and consequent poor performance. Participants were screened and excluded from the study if they had a history of neurological and/or psychiatric disorders (e.g., epilepsy, anxiety, or depression), alcohol and/or drug abuse, head trauma, or surgical implants incompatible with MRI. All participants provided informed consent upon entering the study, which was approved by the Department of Psychology Ethics Board at Swansea University.

Stimuli
Stimuli consisted of colour images of younger and older adult faces, in a frontal orientation. Face images were obtained from the FACES database ( [24]; https://faces.mpdl.mpg.de/imeji/), representing 12 younger (20-30 years of age) and 12 older (60 + years of age) individuals. Each stimulus set was balanced in relation to gender, and stimuli were selected on the basis of prior attractiveness and distinctiveness ratings (use of images with a standard deviation in ratings of less than ten) to ensure that there were no significant differences within and between the different age groups (p > 0.05). Additionally, all selected images featured neutral facial expressions to ensure responses were not confounded by valence [33,65,87].

Experimental procedure
Stimuli were used to create the Go/NoGo task, whereby participants responded each time a face was presented but withheld responses when the same face was displayed in direct succession (see Fig. 1). Each experimental session began with a structural scan (5 min), followed by 2 functional runs of the Go/NoGo task (6.5 min each), one run for younger adult and one run for older adult faces. The order of runs was counterbalanced across participants, as part of an event-related experimental design (which better enables the identification of brain activity unique to inhibition, compared to block designs; [49,57]). Stimuli were presented on a screen positioned behind the MRI scanner and viewed via a mirror mounted onto the head coil. Participants were instructed on how to complete the task before entering the MRI scanner and a reprisal of the instructions was presented for 9 s prior to each experimental run, which advised participants to press a response button with their right index finger each time a stimulus was presented (Go trial), unless the same stimulus was repeated immediately (NoGo trial). Within each run, 240 trials were split into 192 Go trials and 48 NoGo trials, thus representing an 80:20 or 4:1 ratio; sufficient to generate the necessary prepotent tendency for Go responses to facilitate the novelty of NoGo trials [93]. Face stimuli were displayed an equal number of times such that each image featured in 16 Go trials and 4 NoGo trials. Each trial was 1000 ms in length, with an inter-stimulus interval (ISI) of 500 ms. The stimulus remained on screen once participants had responded, such that both stimulus onset and ISI were fixed. As per the aims of the study, trial presentation of this duration, and repeated presentation of a small selection of faces, was implemented to minimise cognitive load. The short, static ISI was adopted to facilitate a consistent and predictable response pattern during the task, in order to produce conditions where it is sufficiently effortful to adequately withhold responses [57,89,99]. Furthermore, varied ISI would alter the length of time participants were expected to maintain representations of the faces in working memory, which would alter subsequent load.
Preprocessing of the obtained images was completed using Statistical Parametric Mapping software (SPM12; http://www.fil.ion.ucl.ac. uk/spm). Functional images were realigned using rigid-body transformation to correct for participant head motion between volumes, and the mean image of each run was co-registered to the structural image. No participants were excluded due to excessive motion (i.e., > 2 mm). The structural image of each participant was then segmented into three tissue types (grey matter, white matter, and cerebrospinal fluid), using tissue probability maps. Spatial normalisation into standard stereotaxic space was completed using the Montreal Neurological Institute (MNI) template with a voxel size of 2 mm 3 . Finally, each volume was spatially smoothed, using a 6 mm FWHM, isotropic Gaussian kernel [17,92].

Analysis of neuroimaging data
The data were analysed with Principal Component Analysis using Partial Least Squares (PLS; [45,52,53]); a multivariate approach, which is optimal for extracting distributed signal changes in relation to task demands. PLS analysis was conducted using a free software (PLSGUI; htt ps://www.rotman-baycrest.on.ca/) implemented via Matlab. PLS reduces the dimensionality of large data sets by decomposing the data into orthogonal dimensions by conducting singular value decomposition (SVD) and outputting a set of latent variables (LVs), i.e., patterns of brain activity related to the experimental design, which account for maximum covariance in the data. This identification of spatiotemporal patterns of whole-brain activity that covary with task effects requires only a single analytical step, as patterns of activity are evaluated across all voxels and timepoints at the same time. Consequently, correction for multiple comparisons is not necessary, which results in higher statistical power than mass-univariate analyses (which consider each voxel separately; [31]).
Here, we used task-based PLS, examining spatial and temporal dependencies among voxels, thus allowing inferences regarding differences across time and space between the experimental conditions. For each condition, we conducted analysis of activity across five TRs, starting at the onset of the face stimulus. Activity at each time point was normalised to the onset TR (i.e., stimulus presentation) and, thus, activity in each condition was uninfluenced by activity in another condition. In an event-related paradigm, PLS provides a set of brain regions related to the experimental conditions for each TR on each LV. At each TR, for each participant, a brain score is calculated by multiplying the salience (i.e., the degree of covariance of activity with the task condition on each LV) of each voxel by the signal of each brain voxel, and summing these across the entire brain. We plotted the mean brain scores at each TR to show overall brain activity fluctuations across the different conditions expressed over the 15 s period, which is analogous to hemodynamic response functions.
To determine statistical significance, we conducted 500 permutation tests to estimate the probability of each LV and 100 bootstraps to estimate the standard errors of the salience for each voxel in order to assess the reliability and robustness of each voxel's contribution to a pattern of brain activity [53]. We used the mean-centering approach, which involves subtracting the grand mean of the data matrix from the task means. We restricted the bootstrap ratio threshold to + /− 3 (statistical significance at p < 0.001; [70]) and reported areas with a cluster size of 50 or more voxels. Confidence intervals (95%) were calculated from the bootstrap; for the mean brain scores in each condition across the five TRs, with significant differences between conditions determined by a lack of overlap in the confidence intervals.
In the current study, we utilised event-related PLS to explore the relations among the following experimental conditions: GO-YF (successful Go trials; younger faces), GO-OF (successful Go trials, older faces), NOGO-YF (successful NoGo trials, younger faces), NOGO-OF (successful NoGo trials, older faces), NOGO-ERR-YF (unsuccessful NoGo trials, younger faces), and NOGO-ERR-OF (unsuccessful NoGo trials, older faces) to examine the presence of an own-age face bias during successful and unsuccessful response inhibition. Specifically, two whole-brain analyses were conducted: first, we examined the relations between the GO and NOGO conditions (GO as baseline), identifying brain activity during successful response inhibition to younger and older faces (n = 44); and, second, we examined the relations between the NOGO and NOGO-ERR conditions (NOGO as baseline), identifying brain activity during unsuccessful response inhibition to younger and older faces (n = 36). For the second analysis, participants with less than four unsuccessful NOGO trials in either face condition were excluded from the analysis.

Behavioural results
To assess performance on the Go/NoGo task, two repeated-measures ANOVAs were conducted: one on the accuracy of responses to successful GO and NOGO trials, and the other on the reaction times to successful GO trials and unsuccessful NOGO trials. For accuracy, the 2 (Trial: Go/ NoGo) x 2 (Stimulus: Younger Faces/Older Faces) ANOVA yielded a significant main effect of Trial (F 1,43 = 66.52, p < 0.001; η p 2 =0.61), demonstrating significantly better performance on the Go trials in comparison to the NoGo trials. The main effect of Stimulus and Trial x Stimulus interaction were not significant (ps > 0.05). For reaction times, the 2 (Response: Successful Go/Unsuccessful NoGo) x 2 (Stimulus: Younger Faces/Older Faces) ANOVA did not reveal any significant main effects or interaction (ps > 0.05; Table 1).

Whole-brain activity: successful response inhibition
The whole-brain analysis of the GO and NOGO conditions (successful trials only) yielded two significant LVs. LV1 accounted for 71.79% of covariance in the data and delineated a pattern of activity common to the NOGO conditions, in contrast to the GO conditions (see Fig. 2). Response inhibition to both younger and older faces engaged a common network of brain regions, including bilateral hippocampus, ventral and dorsal striatum, anterior insula, frontoparietal areas, cuneus, somatosensory cortices, and pre-SMA.
LV2 accounted for 24.78% of covariance in the data and delineated a pattern of activity different for NOGO-YF and NOGO-OF. Successful response inhibition to younger faces engaged orbitofrontal, ventromedial & ventrolateral PFC, bilateral anterior insula, left temporal pole, striatum, right inferior frontal gyrus (BA 9/46), and right temporoparietal junction, whereas successful response inhibition to older faces engaged parahippocampus, left inferior frontal gyrus (BA 45), middle temporal gyrus, precuneus, and pre-SMA (see Fig. 3).

Whole-brain activity: unsuccessful response inhibition
The whole brain analysis of the NOGO and NOGO-ERR conditions yielded two significant LVs. LV1 accounted for 72.32% of covariance in the data and delineated a pattern of activity common to the NOGO-ERR conditions, in contrast to the NOGO conditions (see Fig. 4). Unsuccessful response inhibition to both younger and older faces engaged a common network of brain regions, including bilateral anterior insula, dorsal ACC, intraparietal sulcus, temporoparietal junction, superior temporal gyrus, and pre-SMA.
LV2 accounted for 25.65% of covariance in the data and delineated a pattern of activity different for NOGO-ERR-YF and NOGO-ERR-OF. Unsuccessful response inhibition to younger faces engaged temporal pole, bilateral posterior insula, parahippocampus, thalamus, supramarginal gyrus, premotor cortex, and bilateral precuneus more strongly, whereas unsuccessful response inhibition to older faces resulted in increased activity in caudate nucleus, inferior frontal gyrus (BA 47), angular gyrus, intraparietal sulcus, and superior frontal gyrus (see Fig. 5).

Discussion
A Go/NoGo paradigm, with faces of younger and older adults as stimuli, was used to determine the effects of an age-based in-group bias on the neural responses associated with successful and unsuccessful response inhibition. Behaviourally, we replicated the response inhibition effect, with results showing that participants made more errors in their NoGo compared to their Go responses but did not show any differences when responding to younger or older face stimuli. The neuroimaging results showed sustained activity in regions of the response inhibition network for successful response inhibition (compared to response execution) and in regions of the dorsal salience network for unsuccessful response inhibition (compared to successful response inhibition). Importantly, the results further demonstrated that activity within some regions of these networks was modulated by stimulus type. Thus, the results suggest that age-based in-group bias affects neural information processing during successful response inhibition, as well as during the detection of response inhibition errors.

Activation of the response inhibition network
Replicating previous findings, the neuroimaging analyses revealed that successful suppression of prepotent actions was subserved by activity in the main nodes of the response inhibition network, including bilateral anterior insula, right IFG, MFG, bilateral striatum, and pre-SMA [3,42,56,60,78,80,101]. This sustained brain activity across the functional network was evident for both types of age stimuli, reflecting the response inhibition effect, rather than in/out-group modulation.

In-/out-group modulation of the response inhibition network
The presentation of younger and older faces differentially modulated the activity in several nodes of the response inhibition network, highlighting more specific (regional) rather than network-level differences. In particular, we observed increased vmPFC activity during successful NoGo trials featuring younger faces, compared to older faces. The mPFC is thought to be involved in the detection of socially salient information and has previously been found to display increased activation when viewing faces of a similar age group to the viewer [7,23]. Similarly, our results showed increased activity in the temporoparietal junction, which has been consistently linked to social perception [71] and attention [14]. The findings of the present study, therefore, provide empirical evidence to support the roles of mPFC and TPJ as key structures underlying age-based, in-group processing. In contrast, our results showed that activity was significantly increased during successful response inhibition to older face stimuli within pre-SMA (widely regarded as a core structure underlying motor inhibition, with high connectivity to Fig. 2. Successful response inhibition: shared brain activity across age stimuli. Top: A pattern of whole-brain activity depicting areas active during NOGO vs. GO trials (p < 0.001). Below: Mean brain scores related to the whole-brain activity across four experimental conditions (NOGO-YF, NOGO-OF, GO-YF, and GO-OF). Error bars denote 95% confidence intervals for correlations calculated from the bootstrap procedure; BSR = bootstrap ratio; a.u. = arbitrary units; z coordinate = superior/inferior orientation.
prefrontal as well as motor regions; [58]). This modulation of activity suggests that out-group stimuli require stronger activation of pre-SMA in order to effectively inhibit motor responses.

Activation of the dorsal salience network
The analysis of unsuccessful response inhibition (for stimuli of both age groups) revealed a pattern of activity representative of the dorsal salience network, comprising bilateral anterior insula, dorsal ACC, temporoparietal junction, superior temporal gyrus, and pre-SMA [21,63]. Many of these regions were also found to govern successful response inhibition but it is important to note that a degree of overlap in network activation is to be expected, between trial types, having been produced by the same task [96]. For example, pre-SMA was active in both the response inhibition and salience networks because preparation of a response is characteristic of the general demands of the Go/NoGo task. Similarly, as previously stated, TPJ activation is primarily proposed to reflect the social relevance of stimuli, irrespective of trial type. Activity in regions of dorsal salience network in relation to unsuccessful inhibition is not surprising; especially given the executive role of the anterior insula, shown to execute specialised and integrative functions pertaining to error monitoring [6]. Additionally, while the dorsal region reflects cognitive control processes, the ventral area responds to social and emotional stimulus attributes [85]. This distinction is also evident across hemispheres, with right anterior insula adopting a more affective role; suggesting the bilateral activity observed in the current study is indicative of our stimuli evoking responses requiring the integration of external top-down, cue-dependent information with internal social signals. Likewise, TPJ is also engaged during domain-general salience processing and that involving social cognition, having been shown to be integral to "self and other" discriminations [28,83]. Regarded as a mediator between dorsal and ventral attention networks; an anterior node of TPJ projects to dorsal ACC and MFG, and a posterior node connects to IFG, MFG and precuneusboth of which extend to anterior insula [16,41], highlighting the interplay of these regions in utilising social cues to construct appropriate responses. Fig. 3. Successful response inhibition: differences in brain activity due to age bias. (A) During response inhibition to younger face stimuli (NOGO-YF), significantly more activity was found in right inferior frontal gyrus (BA 9 [maximal divergence at 6 s] and BA 46 [maximal divergence at 3 s] than during response inhibition to older face stimuli (NOGO-OF); (B) During response inhibition to older face stimuli (NOGO-OF), significantly more activity was found in right amygdala (maximal divergence at 3 s) and left inferior frontal gyrus (BA 45: maximal divergence at 3 s) than during response inhibition to younger face stimuli (NOGO-YF). Error bars denote 95% confidence intervals for correlations calculated from the bootstrap procedure; BA = Brodmann area; IFG = inferior frontal gyrus; AMY = amygdala.

In-/out-group modulation of the dorsal salience network
Activity in several nodes of the dorsal salience network was differentially modulated by the presentation of younger and older faces, again highlighting more specific (regional) rather than network-level differences. Unsuccessful response inhibition to younger faces engaged bilateral posterior insula significantly more strongly. Posterior insula projects to parietal and sensorimotor cortices that are instrumental within the ventral system in adjusting responses based on personal attributes [85]. Furthermore, bilateral involvement of precuneus, within superior parietal lobule, also attests to the importance of internal mental representations in response to the stimuli, as this region has a notable role in introspection [10,90]. Therefore, the observed pattern of activity underpinning responses to younger faces indicates that they were regarded as more important, and of higher personal value than those of older adults [23,46,98]. Accordingly, our neural results suggest that it is more salient, or socially detrimental, to make an error in response to faces depicting one's own age group, although this bias is likely to be implicit (given the non-significant behavioural result related to the Fig. 4. Unsuccessful response inhibition: shared brain activity across age stimuli. Top: A pattern of whole-brain activity depicting areas active during NOGO-ERR vs. NOGO trials. Below: Mean brain scores related to the wholebrain activity across four experimental conditions (NOGO-ERR-YF, NOGO-ERR-OF, NOGO-YF, and NOGO-OF). Error bars denote 95% confidence intervals for correlations calculated from the bootstrap procedure; BSR = bootstrap ratio; a.u. = arbitrary units; z coordinate = superior/inferior orientation. stimuli). Socially relevant stimuli, particularly those representing in-vs. out-group distinctions, are therefore highly likely to enhance error awareness and define responses to incorrect trials. For example, the extent of involvement of dorsal ACC has also been linked to the salience of the withholding response [50]. Consequently, age as a stimulus feature should signify a key consideration in future experimental designs relating to response inhibition, and perhaps other executive tasks.
Unsuccessful response inhibition to older faces most notably resulted in increased activity in angular gyrus and intraparietal sulcus. Linked to components of both the ventral (e.g., IFG) and dorsal (e.g., IPS/SFG) attention networks via branches of the superior longitudinal fasciculus [18,51], angular gyrus has been shown to be implicated in conflict resolution and failed inhibition resulting from numerous Go/NoGo tasks [59,76,89]. The activity pattern corresponding to errors involving older faces does not appear to be based on social cognition per se (instead implicating salience purely within the bounds of attention; representing novelty from the perspective of the infrequency of error responses, as opposed to being contingent on the nature of the stimuli). However, the age-related distinction in face stimuli must have been sufficient to evoke the substantial contextual conflict proposed to be required to activate left angular gyrus in this manner (as typically only right hemisphere is shown to be engaged; [73]). Furthermore, connections via the inferior occipitofrontal fasciculus extend from angular gyrus to the caudate nucleus [84], while parahippocampal gyrus is linked to angular gyrus via the inferior longitudinal fasciculus [68]. These structures were also part of the network in question and are instrumental in the chain of events required to integrate perception, recognition, and action control, which angular gyrus is reported to coordinate [60]. Therefore, NoGo errors to stimuli featuring older faces appear to be important in the context of cognitive control but may not be socially meaningful.

Limitations & future directions
With regard to the stimuli, it is important to consider that interpretation of neutral facial expressions may be dependent on the age of the stimulus. For example, younger adults may interpret neutral expressions Fig. 5. Unsuccessful response inhibition: differences in brain activity due to age bias. (A) During unsuccessful response inhibition to younger face stimuli (ERR-YF), significantly more activity was found in bilateral posterior insula [maximal divergence at 3 s] than during unsuccessful response inhibition to older face stimuli (ERR-OF); (B) During unsuccessful response inhibition to older face stimuli (ERR-OF), significantly more activity was found in left angular gyrus [maximal divergence at 6 s] and right intraparietal sulcus [maximal divergence at 3 s] than during unsuccessful response inhibition to younger face stimuli (ERR-YF). Error bars denote 95% confidence intervals for correlations calculated from the bootstrap procedure; pINS = posterior insula; AG = angular gyrus; IPS = intraparietal sulcus.
in older adults as emotional due to the physical changes associated with ageing (e.g., loss of plasticity of the skin, slanting of the eyebrows). While we did not obtain independent ratings of the stimuli in our participant sample, we are confident that the images used would have been interpreted as neutral by the vast majority of participants. This is because the database from which the stimuli were sourced provides accompanying ratings of the facial expressions, which -for the neutral images -were validated in an age-representative sample of 154 participants (comprising equal numbers of younger, middle aged, and older adults; [24]).
Whilst the results attest to an age-bias in neural activation patterns, distinguishing between the processing of younger and older face stimuli in the context of both successful and unsuccessful attempts to inhibit, the lack of behavioural age-bias effect is somewhat surprising. It is speculated that this may have been due to the following. (1) We chose to separate the presentation of younger and older adult faces into distinct functional runs, as opposed to intermixing stimuli, but designed the paradigm in this way for both clarity and brevity. On the subject of clarity, when task responses are based on comparisons and relationships to prior stimuli (as is the case for Go/NoGo paradigms), it becomes essential to block images by defining characteristics, such as age group. As such, intermixing the stimuli would be predicted to alter the basic premise of the task, and may result in participants using bottom-up surface-level age cues to facilitate responses as opposed to engaging top-down inhibition. Previous research has also utilised separate presentations of adult/new born, and adult/child stimuli; generating a statistically significant interaction between the variables (e.g., [46]), suggesting that although it is a possibility, it is unlikely that the manner in which we presented the stimuli contributed to the non-significant own-age bias results. Additionally, in relation to brevity, intermixing the stimuli, while gathering equivalent data, would result in a functional run of ~13 min; demonstrated to induce fatigue and increase the likelihood of motion artefacts [1]. (2) To minimise cognitive load and simplify the Go/NoGo task, the study utilised a small number of face images combined with a relatively slow stimulus presentation rate (1000 ms, compared to 250 ms used elsewhere; e.g., [78]). Nevertheless, it should be noted that the stimulus set size used here is equivalent to that found in other Go/NoGo studies (e.g., those adopting letters stimuli, presenting a total of 12 items; [89]), and across the wider literature investigating the own age-bias (utilising 8-16 images of faces; [2,23]). While we cannot eliminate the possibility of habituation, if habituation were to occur, we would predict that it would influence younger and older adult stimuli in a similar manner (and would, as such, not pose a problem for the purpose of the study). Furthermore, as response inhibition effects were demonstrated behaviourally and neurally, the results suggest that the task was sufficiently challenging to produce errors and distinguish Go from NoGo trials (Crauid & Boulinguez, 2013; [64,99]). Importantly, these elements did not influence the fundamental nature of the task, which evoked activity in response inhibition and salience networks. The fact that age-related distinctions were noted in regions central to action suppression and not in peripheral, face processing regions, also confirms that the paradigm served the intended purpose.
Future research in this area may benefit from the use of skin conductance measures and electroencephalography to further evaluate the effects of error processing [34]. For example, "error-related negativity" (ERN), which is commonly associated with increased autonomic arousal [47,62]. Advancing age is often accompanied by an increased difficulty to recognise faces and engage inhibition mechanisms [12,13,39], such that investigating the responses of older participant samples would also be advantageous in distinguishing whether they use similar or alternative strategies to process out-group stimuli [36,95].

Conclusions
The results of the study provide new insight into patterns of neural responses underlying inhibitory control processes; confirming that both successful and unsuccessful inhibition can be modulated by the stimulus-specific attribute of age and related in-group/out-group biases. Greater activity of mPFC and TPJ to younger faces during successful NoGo trials demonstrates the presence of an own-age bias in the response inhibition network. Furthermore, responses to younger faces during unsuccessful NoGo trials exhibited more prominent involvement of posterior insula and precuneus, indicative of an own-age bias in the dorsal salience network in relation to failed inhibition. Therefore, in generating successful and unsuccessful inhibitory responses, participants appear to have made an implicit appraisal of the age of the stimulus, which was shown to modulate the accompanying network activity. In the context of errors, these findings suggest that young adults found unsuccessful trials involving their own age group particularly salient, possibly attaching a higher degree of value and significance to their processing. Beyond laboratory investigations of response inhibition, this finding has the potential to contribute towards understanding the neural mechanisms of ageism; regarded as a prominent societal concern. As a way of counteracting such implicit bias, training individuals to consciously regulate such introspective processing to view all errors as equal could help to minimise the prospective detrimental impact.

Sources of funding
No external sources of funding. The research was supported by Department of Psychology, Swansea University. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Ethical approval
The study was approved by the Department of Psychology Ethics Board at Swansea University (in line with the Declaration of Helsinki).

Publication ethics
The manuscript adheres to Elsevier's policy.

Declarations of interest
None.

Data Availability
Upon acceptance of the manuscript, the data will be uploaded to a public repository.