**4. Discussion**

Working memory problems and impaired sustained attention are characteristic symptoms of ADHD [2,57]. Improvement of symptomatology by cognitive training and psychological interventions aimed to increase the correlation between sustained attention and arousal has been evalued several times in the recent past [58–61]. However, benefits for behavioral transfer effects to measures of fluid intelligence after several weeks of a computerized working memory treatment requiring high cognitive load could not be confirmed satisfactorily and raised questions about the controversial usefulness of such training [22,24,62]. The demand on cognitive processes is increased by the dual *n*-back task, which is a particular task aimed at challenging the divided attention by running visual and auditory modalities concurrently with the potential of WMT to compensate for a decline in executive functions [47,63]. In the current study, we have tested controls and ADHD patients who were trained during three weeks with the Dual *n*-Back Task. We included two subgroups, from controls and ADHD, who were trained with a non adaptive version of the task-with a fixed level of difficulty set to 1, that is a dual 1-back task (the *baseline* protocol).

We found evidence that WMT, irrespective of *baseline* or adaptive protocol, improved the score of the dual *n*-back task played by the participants at the post-training in comparison with the pre-training session, in agreemen<sup>t</sup> with past studies using fewer sessions of WMT [39]. Our training protocol lasted 20 days, a duration comparable with other studies reported in the literature using Dual *n*-Back Task [64]. In addition to increase in dual *n*-back scores in controls, we observed that WMT improved performance in the WAIS-IV Digit Span in agreemen<sup>t</sup> with previous studies [39,65,66]. It is important to notice that ADHD patients and controls are reliably differentiated by the WAIS-IV [67]. The novelty in our study is that for the first time, to our best knowledge, we report that ADHD patients improved their performance in the WAIS-IV Digit Span after a WM training protocol. After training, ADHD could perform at the same level of controls before training. However, measurement of the visuospatial working memory by the Corsi block-tapping task did not show any significant difference between controls and ADHD neither before nor after training. This finding is in agreemen<sup>t</sup> with the observation that visuospatial working memory is not specifically impaired in ADHD [68–70] and that dual *n*-back task is a working memory task affecting circuits other than those involved in visuospatial processing [66,71–74].

Before training, the results of the probability gambling task did not show any difference between ADHD and controls with respect to the total gain and risk index, in line with our previous study [41]. The analysis of the response time confirmed that ADHD responded at a significant slower pace than controls already before WMT, as previously reported [41]. However, training in both baseline and adaptive conditions provoked a faster reaction in both groups with similar magnitude, thus suggesting a similar process for an increase in the capacity to handle divided attentional stimuli in both ADHD and controls. The WM training failed to affect the total gains, but in the baseline condition it revealed a tendency to increase a risk taking attitude in both groups, matching our previous observation along the same line [75]. Controls tended to increase a risky decision making also after the adaptive training, somehow like after baseline training. On the contrary, after adaptive training, ADHD tended to decrease risk-taking attitude. This result suggests that improved divided attentional processes in both groups and opposite risk-taking behavior are elicited by a high cognitive load generated by the adaptive dual *n*-back task. An interpretation of this result is that these processes are controlled by different pathways, in agreemen<sup>t</sup> with literature on the behavioral deficits of ADHD patients [76–80].

Before the onset of the ERP trigger, we observed a negative readiness potential maximal at frontocentral electrodes, or decision preceding negativity (DPN), consistent with the literature and unaffected by WMT [81]. The ERPs following the gamble selection in the Probability Gambling Task [52] are characterized by several wave components [82–85]. Few studies analyzed the ERPs in adult ADHD and showed that N2, P3a, P3b, feed-back related negativity and N400-like components distinguish ADHD and controls in association with the evaluation of the reward outcome [28,86] and with the emotional feelings generated by risk-taking attitude [41]. In our current paper, we did not observe group effects (i.e., differences between controls and ADHD) on the measurements of PGT due to the working memory training, other than on the response times (Table 2). Our finding is in agreemen<sup>t</sup> with the observation that working memory training in general improves processing speed and attention performance [87] independently of the transfer to fluid intelligence [24,62,79] and that inattentive symptoms in ADHD are not associated with fluid intelligence [78,79,88] Hence, we do not discuss further the gambling task-related ERP components but P1 and LPN, the wave components which appeared to be the most affected by dual *n*-back task working memory training.

We observed a very significant effect of adaptive training on the P1-like wave component in the ADHD group. It is known that P1/N1 early sensory ERP components tend to be attenuated in ADHD patients [89–92] and our observations before training confirm those studies. Attentional modulation progress along the build-up of the ensuing P1-like [93,94]. The attenuation observed in ADHD can be interpreted following the perceptual load theory in selective attention [95,96]. In the current study, the P1-like wave is triggered by the button-click on the selected gamble. ADHD participants might have a degree of perceptual overload when facing the decision to invest, thus impairing their attentional resources as revealed by the attenuated P1-like. A WMT during three weeks with the adaptive training protocol of the dual *n*-back task generates a sustained high cognitive load on divided attention. The particular characteristic of this task is that the working memory capacity is solicited by the number of objects to be memorized and by the cross-modal features associated with the stimuli. P1 was found to likely reflect spatially based information shared by the auditory attention and visual memory systems that do not have to be mutually recruited in situations involving cross-modal tasks [97,98] and sensitive to the number of objects rather to the number of features to be memorized [99]. This is also in agreemen<sup>t</sup> with our finding, mentioned before, of an improvement in the score of the WAIS-IV digit span after WMT. The puzzling finding of an increase in P1-like wave amplitude in ADHD after the adaptive training, with restoration of a waveform similar to controls, is in favor of the hypothesis of an improvement of early higher-level mechanisms of attentional control in ADHD after adaptive training. The topological maps have clearly located this change of activity at the level of the prefrontal cortical areas. The P1 wave has been associated with an inhibitory feedback wave from "higher" cortical areas acting as an inhibitory filter to control feedforward sensory processes [100]. A change in P1 might be related to a change in the early modulation of attention, such to improve the sensory-perceptual level of processing that is necessary to improve the decisional process. There is evidence that modulation of neural activity by selective attention may occur at the subcortical level [101], where inhibitory gating mechanisms take place [102–104]. Then, an increase of the P1-like wave, in our paradigm, could be associated with a more effective processing of the decision due to a greater inhibition of potentially competing and task irrelevant networks.

The last ERP component strongly affected by WMT is the late parietal negative slow wave, whose amplitude was selectively increased in controls after both training level conditions. This wave started approximately 700 ms after the onset and extended for several hundreds of milliseconds. The topographic distribution of this wave is clearly parietal-posterior and with a left hemispheric dominance at latencies near 900 ms. In the literature, it is interesting to note that LPN has been observed as a neural marker related to the transfer of cross-modal associated information in working memory, [105,106] with memory tasks that required continued evaluation of contextual information [107–109] and with memory tasks that required high demands on action monitoring in presence of conflictual response options [110]. Before training, ERPs recorded in both groups did not show a relevant presence of LPN in the PGT. This suggests that WMT is a necessary condition to let appear LPN. Moreover, ADHD patients did not show any effect of WMT on LPN. If we consider all these observations together we may raise the hypothesis that WMT with the dual *n*-back task is able to generate a transfer effect in the PGT [111,112], as revealed by the LPN associated with a continued evaluation of contextual information in our PGT. This effect is strong in controls, but it is absent in our ADHD group, thus suggesting that the kind of WMT performed by our group of patients only confers benefits for those tasks that were trained [15,62,113]. We could speculate that a deficit in transfer effects associated with WM training could be associated with the abnormal parietal brain function observed in ADHD [114].
