Nineteen children with AS and 19 control individuals participated in this study. Individuals in the AS group were selected from the outpatient population of the Institute of Cognitive Neurology (INECO) and were assessed by a psychiatrist. Their diagnosis was based on the criteria established by Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) (American Psychiatric Association, 2000). Additionally, the patients' symptom presentation was measured using the Autism Quotient (AQ) for children (Baron-Cohen et al., 2001b) and adolescents (Baron-Cohen et al., 2006). This questionnaire includes traits of autistic patients which are overlooked in other diagnostic tools (Baron-Cohen et al., 2001b, 2006; Auyeung et al., 2008). We also employed the Social Communication Questionnaire (SCQ; Bolte et al., 2008), which is based on the Autism Diagnostic Interview-Revised, and is widely used in clinical research and practice (Chandler et al., 2007; Norris and Lecavalier, 2010). A psychiatrist then validated the symptom examples provided by the AQ and SCQ and checked the other AS symptoms and criteria.

Twenty-two typically developing children were recruited from neighboring schools. Nineteen of these participants were selected to form a control group, matched for age, gender, and fluid/crystallized intelligence with respect to the AS group. Note that both measures of intelligence were used as matching criteria, so as to prevent the underestimation of intelligence by employing a single criterion (Hayashi et al., 2008; Soulieres et al., 2011). Moreover, given that AS children obtained high variability in fluid intelligence and low variance in crystallized intelligence (see SDs in Table 1), control-participant selection was based on group-wise rather than pair-wise matching criteria. The groups showed no significant differences on any of the matching measures (see Table 1). The following exclusion criteria for both groups were applied: (1) participants who met DSM-IV criteria for any axis-I; and (2) individuals with a history of intellectual disability, neurological disease, psychiatric disease (except AS in patient group), or any clinical condition that may affect cognitive performance.

Parental written informed consent was obtained in accordance with the declaration of Helsinki. The study was approved by the ethics committee of the INECO.

An extended battery of neuropsychological tests was used to assess cognitive functioning, including measures of intelligence, motor speed, memory, visuo-spatial constructional ability, EFs, and ToM.

Group differences were analyzed through an ANCOVA test using age as a covariate. Eta squared (n²) was employed as a measure of effect size for the significant effects. In addition, we included an inferential test used to assess variance equality between two groups (only significant differences were reported). To further assess inter-individual differences, we conducted MCSA and compared each participant with the mean of the control group on every measure. We followed the method of Towgood et al. (2009) by using a threshold of two standard deviations (SDs) from the mean of the control group to define the normal range. First, we removed control children who displayed extreme performance in each sub-measure, according to the two SD criteria. Second, we recomputed the control means and SDs excluding these subjects and identified AS and control participants whose performance was sub-normal (two SDs below control mean), supra-normal (two SDs above control mean), and average (between −1.99 and 1.99 SDs from the control mean). Third, the participants previously excluded were re-included for MCSA (see Figures 2A,B).

We then used non-parametric (Kruskal-Wallis and Mann-Whitney) tests to assess whether the number of measures in which AS individuals obtained sub- and supra-normal performance was associated with individual differences in FI. Finally, we used Spearman's rank correlations to examine the association between FI and neuropsychological measures. The significance of all correlations was corrected for multiple comparisons using the Sidak method. The adjusted α level after correction was set at 0.002. The α value for all other statistical tests (not related to correlation) was set at 0.05.

Table 2 shows the significance of group comparisons after ANCOVA, using age as a covariate. Following covariation, the AS group was significantly impaired on verbal memory acquisition [RAVLT Acquisition: F_{(1, 35)} = 5.72, p = 0.024, n² = 0.175], visuo-spatial constructional ability [RCFT copy: F_{(1, 35)} = 16.60, p = 0.000, n² = 0.335], and ToM [RMET: F_{(1, 36)} = 6.45, p = 0.016, n² = 0.164] (see Figure 1A). For these differences, age was significantly related to the RCFT [F_{(1, 35)} = 17.34, p = 0.000, n² = 0.344], but not to RAVLT acquisition [F_{(1, 35)} = 1.22, p = 0.278, n² = 0.043] or RMET [F_{(1, 35)} = 0.277, p = 0.602, n² = 0.008].

Although participants with AS presented lower scores on receptive vocabulary (vocabulary from WISC III), working memory (digit span and arithmetic from WISC III), motor processing speed (coding and symbol search from WISC III), cognitive flexibility (TMT-B), response inhibition (Stroop task), and logical memory (story memory from WRAML), no significant differences were observed in these measures. However, the SDs of some measures were higher in the AS group (see Table 2). A test comparing the group variance revealed significantly higher SDs for the AS than control participants in the RPMT (p = 0.000), RCFT copy (p = 0.016), and story memory immediate recall (p = 0.001). Consistent with previous reports (Hill and Bird, 2006; Towgood et al., 2009; Baez et al., 2012; Gonzalez-Gadea et al., 2013), these data show a consistent pattern of high inter-individual performance variability in AS.

First, we explored performance variability among AS children (Towgood et al., 2009). For each group, in every measure, we calculated the percentage of sub- and supra-normal performers and the percentage of outliers (sub-normal plus supra-normal performance). In the control group, the maximum percentage of outliers was 11% (see Figure 2B). Regarding the AS group, 10 out of 21 measures exceeded this maximum percentage.

In this group, some measures revealed only sub-normal performance: the copy of RCFT, with 32% of sub-normal performers; and TMT-B, vocabulary from WISC-III, and RMET, with 21% sub-normal performers. However, the highest proportion of outliers was observed in tasks where individuals obtained both sub- and supra-normal performance. Thus, the RPMT exhibited the highest proportion of participants with extreme performance (47%), followed by SM immediate recall (37%), coding from WISC III (32%), and arithmetic (27%). There were no participants who performed only supra-normally.

Therefore, we used the Kruskal-Wallis test to compare the number of measures with sub- and supra-normal performance according to the patients' FI score. Following Towgood et al. (2009), we separately recounted the number of measures (except FI) in which each AS individual obtained sub and supra-normal performance. Additionally, we used FI to categorize three groups of participants with inferior (<2 SDs), superior (>2 SDs), and average (between −1.99 and 1.99 SDs) scores. Table 3 shows that significant group differences were observed only in the number of measures in which participants obtained sub-normal performance (H = 8.37, p = 0.015). After that, we used Mann-Whitney tests for pair-wise comparisons. Participants with superior FI scores displayed a smaller number of measures with sub-normal performance than children with inferior (U = 1.00, p = 0.019) and average (U = 4.00, p = 0.007) FI scores. No significant differences between AS children with average and inferior FI (U = 16.00, p = 0.558) were observed. These results suggest that AS children with higher FI have a lower probability of showing deficiencies in other domains.

Finally, to further explore the influence of FI on other domains, we conducted a correlation analysis between FI (RPMT) and neuropsychological measures in each group (see Table 2).

For the AS group, there was a significant association between FI and cognitive flexibility (TMT-B: r_s = −0.72, p = 0.002; TMT Interference: r_s = −0.80, p = 0.000). A significant correlation was also found between FI and motor processing speed (coding: r_s = 0.83, p = 0.000; symbol search: r_s = 0.73, p = 0.001) (see Figure 3). For the control group, we found no significant correlation between FI and neuropsychological tasks.

In summary, MCSA revealed that group comparison analyses are blind to the heterogeneity in children with AS. Furthermore, AS individuals with higher scores on FI evidenced fewer difficulties in other cognitive domains and improved performance in cognitive flexibility and processing speed tasks.

The results demonstrated significant group differences between AS and control children in RMET. This suggests that AS children have difficulties in inferring the mental states of others. However, the MCSA showed that only a few children of the AS sample performed sub-normally, confirming that group differences are not always a reliable index of deficits in the AS group. Furthermore, as in the case of AS adults (Ponnet et al., 2004; Spek et al., 2010; Baez et al., 2012; Gonzalez-Gadea et al., 2013), the RMET might not be a sensitive instrument for detecting ToM abnormalities in AS children.

In line with previous evidence of global processing deficits in autism (Minshew and Goldstein, 1998), our AS group showed failures in visuo-spatial constructional abilities (RCFT) and memory/learning acquisition (RALVT). Other studies using visuo-spatial tasks have shown that individuals with AS tend to focus on details rather than global figures, a strategy that is ineffective for RCFT performance (Prior and Hoffmann, 1990; Mottron and Belleville, 1993; Le Sourn-Bissaoui et al., 2011). Since this task requires information encoding supported by organizing and planning strategies (Watanabe et al., 2005; Ogino et al., 2009), difficulties in EFs may underlie these results. In the same vein, when participants are instructed to learn a list of words, inefficient encoding strategies would imply lower acquisition and recall of information (Minshew and Goldstein, 2001; Bowler et al., 2008). Patients with EF impairment perform poorly on unstructured word-list memory tasks, but not on logical memory tests (Tremont et al., 2000; Brooks et al., 2006; Torralva et al., 2011). Our AS participants exhibited difficulties in RAVLT acquisition but not in logical memory tests (story memory). The discrepancy between the memory tests may reflect executive difficulties in organizing efficient information encoding strategies. However, MCSA revealed that six children with AS exhibited inferior performance in the RCFT, while only two children obtained sub-normal scores in the RAVLT. This suggests that difficulties in global processing tasks are also heterogeneous between AS children.

In addition, as was the case in previous studies (Ambery et al., 2006; Hill and Bird, 2006; Nyden et al., 2010), we found no significant differences between groups on the EF tasks However, we did find marked inter-individual variability in these functions, which may account for the absence of group differences. In this sense, Liss et al. (2001) suggested that the problem of universality in executive dysfunction in ASD is that most studies focus on group differences, neglecting individual variations. Our results support this view.

Various authors agree that research on AS should abandon the search for a single cause and address it as a complex, multifactorial syndrome (Happe et al., 2006b; Willcutt et al., 2008; Brunsdon and Happe, 2014). In our study, performance on cognitive measures was more heterogeneous in AS than control children (see Figure 1B). The patients showed extreme performance, including sub-normal (<2 SD), supra-normal (>2 SD), and combined scores. Consistent with our findings, previous reports showed similar patterns among adults with AS (Towgood et al., 2009; Baez et al., 2012; Gonzalez-Gadea et al., 2013). Ours is the first study to confirm this cognitive profile in children with AS.

MCSA of the AS group revealed that several patients presented sub-normal performance in domains that may be associated with their diagnostic categories, such as social interaction (ToM), verbal communication (receptive vocabulary), and repetitive interests and activities (cognitive flexibility). Moreover, the group exhibited both sub- and supra-normal performance in domains excluded from diagnostic criteria, such as information processing styles (logical memory test and RCFT), processing speed (coding from WISC III), and working memory (arithmetic from WISC III).

In support of our first hypothesis, subtle differences across cognitive domains were not revealed by the group-type analysis. Thus, in domains associated with the AS diagnosis, this group included a high proportion of individuals with sub-normal performance, while in other domains, AS children obtained either sub- or supra-normal performance.

For instance, although AS and control groups were similar regarding FI, the former showed high inter-individual variability on the RPMT. Indeed, this group displayed the greatest variability in FI. Most previous reports described FI as either an intact or superior ability in individuals with AS (Hayashi et al., 2008; Chen et al., 2010; Morsanyi and Holyoak, 2010; Soulieres et al., 2011). However, these studies failed to mention the patients' strong variability in the RPMT. For example, in the study by Soulieres et al. (2011), the SDs of the AS group on this task were more than two times larger than in the control group.

Since FI may be understood as a general intelligence factor that contributes to all cognitive functions (Spearman, 1904), we hypothesized that individual differences in FI should affect the cognitive profile of AS participants. We found significant differences between participants with high, low, or average FI in terms of the number of cognitive measures with sub-normal performance. Participants with superior FI demonstrated less impairment in other cognitive functions. Similarly, most executive deficits in patients with frontal dysfunctions are explained by a loss in FI (Duncan et al., 1995; Roca et al., 2010; Woolgar et al., 2010). Likewise, improved FI performance is related to better psychosocial adaptation in typically developing children (Huepe et al., 2011). Our results suggest that high FI may reduce AS children's vulnerability to develop deficits in other cognitive functions.

In view of previous findings, we investigated the association between FI and other cognitive functions in both AS and control groups. Correlation analyses revealed that FI was linked to cognitive flexibility and processing speed only for children with AS. Our data are consistent with recent studies suggesting that FI is a substantial contributor to classical EF tasks (such as TMT) in neurological patients (Roca et al., 2010, 2012). However, this effect cannot be attributed to a positive correlation between FI and all frontal functions. In the present study, ToM was not correlated with FI, as reported elsewhere in the literature (Roca et al., 2010, 2012).

Previous reports indirectly support our findings. First, Kaland et al. (2008a) suggested a link between low FI and limited cognitive skills to solve visuo-constructional problems in a group of children with AS. Second, better attention switching has been shown to predict a higher RPMT overall score in typically developing individuals with variants of the autistic phenotype (Fugard et al., 2011). Finally, Soulieres et al. (2011) suggested that high FI in children with AS would provide them with better mechanisms to solve cognitive tasks.

Finally, the relationship between FI and EFs has been supported by functional magnetic resonance imaging (fMRI) studies. The lateral prefrontal cortex and posterior parietal regions are the neural substrates subserving the relation between abstract reasoning (FI) and performance on EF tasks among neurotypical adults (Gray et al., 2003; Lee et al., 2006; Woolgar et al., 2010). For their own part, participants with AS exhibit significantly increased activation in the lateral prefrontal cortex and left parietal brain regions during EF tasks (Schmitz et al., 2006). Increased brain activation has been explained by abnormal brain anatomy or the use of alternative cognitive strategies (but see opposed results in autism: Luna et al., 2002; Koshino et al., 2005; Soulieres et al., 2009).

By combining evidence from MCSA and correlation analysis, we offer preliminary evidence that AS children with strengths in FI may develop efficient strategies to perform some cognitive tasks. However, the neural mechanisms underlying FI and cognitive functioning in individuals with an AS diagnosis remain unknown. Futures studies should investigate the neural networks associated with this interaction.

The present data confirm heterogeneous cognitive profiles in classical neuropsychological tests among children with AS. However, the origins of this variability remain unknown. Furthermore, previous studies have demonstrated that variation in social cognition tasks is a feature of AS adults (Baez et al., 2012; Gonzalez-Gadea et al., 2013). Future research should explore the issue of heterogeneity in social cognition among AS children.

This study provides preliminary evidence indicating that individual differences in FI may be associated with the heterogeneous profile of strengths and weaknesses of children with AS. Nevertheless, the results should be extended and replicated. The limited sample size in this study did not allow us to establish a definitive causal relationship between FI and heterogeneous cognitive profiles. Regression analyses and structural equation modeling (SEM), which require an extended number of participants, could be robust models to predict the role of FI in cognitive profiles.

Furthermore, the observed heterogeneity may allow for alternative explanations. For instance, clinical measures of autistic symptomatology have been associated with cognitive heterogeneity in adults with AS (Hill and Bird, 2006). Future studies should replicate these findings in children with AS. Finally, traditional approaches interpret variability as noise in the data, which was attributed to limitations of the AS diagnosis, leading to multiple subgroups of patients (Towgood et al., 2009). The DSM-5 has incorporated the AS diagnosis within ASD, in an attempt to account for the variations in symptoms and the multifaceted cognitive profile of each patient (Brunsdon and Happe, 2014). Future studies should investigate the sensitivity of this new diagnosis to account for the heterogeneous cognitive profile of these children.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.