**4. Discussion**

In the present study, preschool children's performance on VSSAT replicated previous behavioral results [38]; their concurrent ERPs showed that, for most electrodes, the amplitudes were more pronounced in target than distractor trials from about 50 ms to 650 ms post-stimulus presentation. However, response to distractor had relatively higher amplitude than targets in right temporal and left parietal electrodes in the interval ranges of 200–400 ms and in the midfrontal electrode at 500–700 ms. These exceptions concerned a narrow subset of electrodes, for a narrow time range, with relatively smaller effects, and might be interpreted as correlates of *interference control*, namely, resistance to interference from the distractor and suppression of its impact on ongoing selective-set processing aimed at releasing or inhibiting the appropriate (correct) manual response [68].

In the task we used, the ERP correlates of enhancement were confounded with those of the execution of manual response only after 400 ms. The determination of response times within ±2 standard deviations from the grand mean (685 ms) reveals that manual response in most cases could be estimated to occur between 425 and 945 ms. Therefore, selective-set effects were unconfounded by manual motor processing across bin-intervals up to 400 ms. Within this time interval, the peak analysis showed evidence of enhancement of ERP to target except for the two instances mentioned. This analysis also corroborated older findings in showing anteriorly and centrally distributed adult-like N200 and N400 and posteriorly distributed P100 and P300 waveforms. Therefore, we conclude that the present findings support a slightly amended version of the enhancement hypothesis in that a parallel relatively minor and segregated processing may have occurred in a subnetwork in order to suppress the distractor's interference.

By triangulating response times analysis, peak analysis, source dipole modeling, simulation, and spiking modeling, we provided converging evidence separating the selective-set processes as occurring earlier (before 400 ms) than the actual response (at around 700 ms, on average) and showing ye<sup>t</sup> another distinct spatiotemporal pattern of activity from those associated with later processes (between 500 and 650 ms), which presumably reflected the planning and preparation of response.

We also performed activity paths analysis to illustrate plausible sequence of the prominent neural activity over time for target and distractor based on maximum differences in peak amplitudes. The results seem compatible with the interpretation that attending the target while holding in working memory the response plan yielded a neural path starting from the parietal regions, then to right temporal and central regions and finally to the frontal regions. In contrast, withholding responses to distractor seemed associated with punctuated activity at the left parietal and right temporal regions. Importantly, comparison of neural activity between target and distractor reveals that in the initial 400 ms there was less involvement of frontal areas concurrent to the distractor.

Dipole analysis further suggests possible and plausible neurofunctional pathways and dynamics involved at cortical and subcortical level. This analysis (see Left Panels of Figures 7 and 8) estimated that initially, within the first 150 ms, the generators of scalp ERP signals, which are distributed posteriorly at right and mid parietal electrodes, seem to correspond to activity in the precuneus. Subsequently, around 250 ms, the ERP activity, mapped on the scalp at left temporal, central, and right frontal electrodes, seems to correspond to a dipole source in the left supramarginal gyrus of the left temporal cortex. Before what might be reasonably considered the timeframe for implementing the manual response, around 350 ms, ERP activity mapped at mid parietal and frontal electrodes was estimated as being generated at the level of a dipole located in middle frontal gyrus of the right frontal lobe. At approximately 45 ms, ERP activity was then estimated to a source in the right insula. The following ERP activity, at 550 and 650 ms, was associated with dipoles in the medial dorsal nucleus of the right thalamus and the right caudate body, respectively.

In contrast, ERP activity concurrent to the distractor was accounted for by just two dipoles, the first occurring at 250 ms in a source located in the middle temporal gyrus, at the junction of right occipital and temporal cortices, the second occurring at 550 ms in the left claustrum.

The previous analysis converges with the findings of a close fit between the preschool children's ERP topographic as well as derived source data and the adult-based ACT–R functional and structural simulations. This convergence suggests evidence of neurofunctional homology. In addition to providing preliminary validity and reliability to the analysis based on the actual data, the converging results from the ACT–R modeling may provide a framework for interpreting the results in a coherent meaningful way and for a detailed inferential reconstruction of the plausible, possible underlying neurocognitive processes related to the enhancement mechanism.

As modeled through ACT–R, our findings may be interpreted as showing that in the initial phase of the VSSAT, target presentation might have been associated with the involvement of key structures of the dorsal (precuneus, BA 7) and ventral (supramarginal gyrus, BA 40) attentional networks [69–71]. These structures appeared to be activated relatively early, similarly than in adults. Next, activation seemed to follow in two functionally interconnected parts, one in the dorsolateral prefrontal cortex (middle frontal gyrus, BA 10); the other in the frontal part of the dorsal attentional network (insula, BA 13). The literature indicates that these structures seem to be activated during the processing of deviants and standards, and specifically, both structures seem to be involved in voluntary target detection, playing important roles in top-down selective attentional control [72–74]. Successively, activation seems to have involved the thalamic dorsomedial nucleus, which might play a role in the regulation of cortical networks, especially when the maintenance and temporal extension of persistent activity patterns in the frontal lobe areas are required, as in the case of sustained attention [75]. The final stage leading to manual response seemed to be associated with the involvement of a key structure in basal ganglia-striatum network, the caudate body. This "cognitive" part of the caudate seems to participate in the control of action including executive functions such as working memory, set shifting, and inhibitory control [76–78].

The presentation of the distractor seemed to be associated with early engagemen<sup>t</sup> of the middle temporal gyrus (BA 37) which is generally deemed to be involved in visual recognition and verbal labeling/categorization [79]. Subsequent late activation seemed to involve the claustrum, a structure thought to participate in the regulation of vigilance and in voluntary allocation of attention [80].

Triangulation of peak analysis, ACT–R modeling and simulation for the entire ERP epochs up to the moment of manual response (~700 ms, on average) suggested converging evidence of distinct but separate interacting processes of enhancement and planning for response release/inhibition, respectively. Thus, the results from the triangulation, considering both target and distractor conditions, overall sugges<sup>t</sup> potentially important interrelations between basal–parietal–temporal–frontal–basal loops and large-scale attentional networks. The feedback loops and functional connectivity originating and ending in the basal ganglia and the striatum, as postulated by the ACT–R architecture, control diverse behaviors largely but not exclusively involved in high-level perception, walking, talking, thinking, language, comprehension, associated with frontal lobes, where the motor strip also sits [81]. As some theories have proposed [82], the aspects which might undergo fine-tuning in young children, in terms of speed and e fficiency, most likely may not involve the selective-set in isolation, but rather its voluntary and flexible coordination with high-level perceptual and working memory processes, recruited for selecting, monitoring and executing or inhibiting the appropriate behavioral response.

A brief discussion of the main limitations and caveats of our study is in order. While the number of electrodes and montage set-up we used may be justified for dipole source analysis in young children (as we have detailed in the methods section and especially for practical challenges in collecting EEGs in this population), it may not be sufficient for reliable source analysis in adults. Therefore, the results need to be replicated in further studies which combine real fMRI and possibly eye movement measurements. Furthermore, as we have already noted ACT–R simulation could be refined (or even replaced by a more flexible architecture) to improve fit and predictions in timings and anatomical localization of neurofunctional modules.

Lastly but not least, although our modeling and simulation procedures are grounded in the literature, being validated by separate tests of the most relevant components of ERPs from actual adult samples [83–85], still we did not report data from an actual adult comparison group. While we acknowledge that in this respect our results are preliminary, and indeed this is a desirable priority for future research, adult comparison on this version of the VSSAT might not necessarily augmen<sup>t</sup> the strength of the supporting evidence because it presents non-trivial methodological challenges. In particular, we have learned from small pilot studies in our lab that the current VSSAT is not appropriate for adults. This task needs to be adapted to prevent confounds of other spurious aspects (i.e., boredom, engagemen<sup>t</sup> level, ceiling effects) occurring in adult participants (but not in preschool children). In other words, this task should be modified significantly to "equate" children's state and age-appropriate task demands. However, as pointed out by others (notably, see [17]), if children and adults are not compared on the same task, data interpretation would still depend on assumptions derived from a priori hypothetical models. Consequently, under such differing conditions, the correspondence defining homology on the basis of comparisons between actual adult and children ERP data would still be based on a type of model-mediated inductive inference. This would be essentially similar to the present approach, therefore, the resulting evidence would not be more "direct" or "realistic" or logically different than the one offered here.
