The corpus callosum as anatomical marker of intelligence? A critical examination in a large-scale developmental study

Intellectual and general cognitive abilities are supported by a large-scale brain network encompassing association cortices in frontal, parietal, and temporal lobes (Deary 2012; Deary et al. 2010; Jung and Haier 2007; Luders et al. 2009; Shaw 2007). The nodes constituting this network are distributed across both cerebral hemispheres, emphasizing the relevance of functional interaction between the hemispheres for performance in tasks demanding higher intellectual abilities (e.g., Belger and Banich 1998; Davis and Cabeza 2015; Welcome and Chiarello 2008, for review see Banich 2003). Consequently, it has long been suggested that the strength of the commissural connections—e.g., reflected in the midsagittal size of the corpus callosum—might serve as an anatomical marker of higher intellectual abilities (Hulshoff Pol et al. 2006; Men et al. 2014; Spitzka 1907; Strauss et al. 1994). A larger corpus callosum and/or thicker myelinated callosal axons would improve inter-hemispheric connectivity and, in turn, intellectual performance. In line with this interpretation, a common genetic origin for corpus callosum size and intelligence has been suggested in studies on healthy twins and siblings (Hulshoff Pol et al. 2006). This general notion is supported by a series of recent magnetic resonance imaging (MRI) studies demonstrating correlations between intelligence coefficients (IQ; as measured with standard intelligence tests) and measures of mid-sagittal callosal area (e.g., Allin et al. 2007; Ganjavi et al. 2011; Hutchinson et al. 2009; Peterson et al. 2001), thickness (Luders et al. 2007, 2011), and microstructural integrity (i.e., fractional anisotropy; Chiang et al. 2009; Dunst et al. 2014; Hutchinson et al. 2009; Navas-Sanchez et al. 2014; Tang et al. 2010). These studies also reveal an apparent dissociation in the direction of the reported correlations as a function of age. Studies examining adult samples mostly report positive associations (Chiang et al. 2009; Dunst et al. 2014; Luders et al. 2007; Strauss et al. 1994; and selectively in female participants in Tang et al. 2010). On the other hand, studies examining children, adolescents, and young adults report negative associations (Allin et al. 2007; Ganjavi et al. 2011; Hutchinson et al. 2009; Luders et al. 2011; but see Nosarti et al. 2004). For example, Luders et al. (2007) studying a mostly adult sample (age range 16–44 years) observed significant positive correlations between IQ measures and thickness in the posterior half of the corpus callosum. Studying a developing sample (age range 6–17 years) with the same methodological approach, the same group found a significant negative correlations, again located in the posterior corpus callosum (Luders et al. 2011). It has been suggested that this developmental dissociation might be attributed to differences in the relative weight of intra- vs. inter-hemispheric processing (Ganjavi et al. 2011) and age-related changes in task demands (Hutchinson et al. 2009).

When interpreting these previous findings it also has to be considered that in all the above studies intelligence was quantified using age-standardized, norm-deviation IQ scores. These deviation IQ scores reflect the relative position within the norm group, and do not represent the absolute level of performance (Angoff 1984; Neisser 1997; Wechsler 1999). As a consequence differences in performance levels between norm groups (i.e., between participants converted with different conversion tables) are removed and set to the norm distribution’s mean (usually 100; see e.g., Angoff 1984; Neisser 1997). The latter effect of the IQ conversion is especially pronounced when studying development samples since the provided conversion tables for children and adolescents cover very narrow age spans reflecting the rapid absolute intellectual development in this period of life. For example, the Wechsler Abbreviated Scale of Intelligence (WASI, Wechsler 1999), which was used by all of the above-mentioned developmental studies, provides conversion tables covering age spans of only 4 months for participants under the age of 16 year. As a result, the mean of the converted IQ scores will appear stable throughout childhood and adolescence (e.g., Burgaleta et al. 2014) although the absolute level of performance (e.g., the mean raw test scores) rises continuously in this period of life (Neisser 1997; Tamnes et al. 2010). While the conversion to IQ score allows evaluating age-appropriateness of an individual’s intellectual abilities and rank stability within a cohort over time, only the use of raw test score will allow to study the development of intellectual functioning. Therefore, we here argue that to address the question of whether corpus-callosum morphology can serve as marker of intellectual and cognitive abilities during development, measures of absolute rather than norm-relative performance should be used.

The present study was designed to systematically re-examine the association of IQ test measures and corpus-callosum anatomy during development in a large mixed cross-sectional and longitudinal sample (734 datasets) by for the first time using raw test scores rather than deviation IQ. This was done in three analyses steps. First, to establish a general structure–function association, we related verbal raw test scores (v-RS) and performance raw test scores (p-RS) to regional callosal thickness measures. As brain size has previously been found to be positively correlated with both corpus callosum size (Ganjavi et al. 2011; Jäncke et al. 1999; Westerhausen et al. 2016) and intelligence measures (for a meta-analyses see McDaniel 2005; or Pietschnig et al. 2015) total intracranial volume (TIV) was included as an additional predictor in a second analysis step. Finally, in a third step, we examined the effect of chronological age on the structure–function association. This was necessary as raw intelligence test scores (Neisser 1997; Tamnes et al. 2010) and callosal thickness (e.g., Giedd et al. 1996; Luders et al. 2010b; Westerhausen et al. 2016, 2011) are characterized by a continuous age-related increase in the studied age period, and the temporal co-occurrence of these two developmental trends could confound the association (see e.g. Salthouse 2011).

The analysis 1 for both p-RS and v-RS revealed a significant positive association of test performance and callosal thickness in the posterior corpus callosum including most parts of the splenium section (see upper row, left panel, Figs. 2, 3). For p-RS the strongest association was found in segment 58 with an explained variance of ω ² = 0.07 (β _p-RS = 1.00; t ₇₂₈ = 5.01, p < 0.0001, see Fig. 4). Comparably, for v-RS the strongest association was also found in segment 57 (β _v-RS = 1.01; t ₇₂₈ = 5.03, p < 0.0001; ω ² = 0.07, see Fig. 4). The v-RS analysis additionally revealed a negative association in the genu of the corpus callosum, that is, in segment 13 (β _v-RS = −0.55; t ₇₂₈ = −3.14, p = 0.0018, ω ² = 0.03, see Fig. 2). In addition to the main effect, both analyses also revealed a significant interaction of test score and Sex indicating differences in the slope of the association between the sexes. In the p-RS analysis the interaction was found in extended areas in the genu as well as in the truncus region with the maximum effect being located in segment 9 (β _p-RS*Sex = 1.44; t ₇₂₈ = 4.12, p < 0.0001, ω ² = 0.05). Across all significant segments, the interaction was driven by a more positive slope in female than in male participants (see Fig. 4d). In the v-RS analysis the area of significant interactions was restricted to the genu region with the maximum being located in segment 12 (β _v-RS*Sex = 1.09; t ₇₂₈ = 4.17, p < 0.0001, ω ² = 0.05). Also here the interaction was driven by a more positive slope in female than male participants (see Fig. 4b).

In analysis step 2, including TIV as covariate, for both intelligence measures a positive association with thickness in posterior callosal segments was found (see Figs. 2, 3, second row). Comparable to analysis step 1, the strongest association was located in segment 57 for the v-RS analysis (β _v-RS = 1.13; t ₇₂₇ = 5.54, p < 0.0001; ω ² = 0.09, see Fig. 5) and in segment 58 for the p-RS analysis (for p-RS: β _p-RS = 1.07; t ₇₂₇ = 5.36, p < 0.0001; ω ² = 0.08). However, including TIV in the model, the interaction of Sex and Test Score was no longer significant (for v-RS all |β _v-RS*Sex| < 0.51; all |t ₇₂₇| < 1.66; and for p-RS all |β _p-RS*Sex| < 0.69; all |t ₇₂₇| < 2.08; all non-significant using FDR correction).

Finally, including Age covariates (i.e., linear term, quadratic term, interaction of test score and Age) in analysis step 3, the main effect of test score vanished for both test subscores (for v-RS all |β _v-RS| < 1.04, all |t ₇₂₄| < 2.41; for p-RS all |β _p-RS| < 0.76, all |t ₇₂₄| < 2.00; all non-significant using FDR correction) for both p-RS and v-RS analyses. In addition, the interaction of test score and Sex did not reach significance (for v-RS all |β _v-RS*Sex | < 0.47, all |t ₇₂₄| < 1.53; and for p-RS all |β _p-RS*Sex| < 0.70, all |t ₇₂₄| < 2.11). In neither of the two analyses a significant interaction of Test Score and Age was found (for v-RS all |β _v-RS*Age| < 0.20, all |t ₇₂₄| < 2.72; and for p-RS all |β _p-RS*Age| < 0.22, all |t ₇₂₄| < 2.52; all non-significant using FDR correction, see Fig. 6), that is, the association of intelligence and callosal thickness was not significantly modulated by Age.

The present study re-examined the relationship between intellectual development and callosal maturation during childhood, adolescence, and young adulthood. When not accounting for chronological age, a positive association in the splenium of the corpus callosum was detected for both p-RS and v-RS. At the thickness segments with the largest effect size, an increase in p-RS or v-RS by one standard deviation was reflected by an increase in regional callosal thickness of about 1 mm. This positive association was found irrespective of whether individual differences in brain size were accounted for in the analysis or not. Considering the known topographical organization of the corpus callosum, the splenium region interconnects temporal, parietal, and occipital cortices, including sensory and higher association areas (Putnam et al. 2010; Schmahmann and Pandya 2006; Westerhausen et al. 2009). Temporal, parietal, and occipital regions have shown associations to IQ scores in developing (Burgaleta et al. 2014; Karama et al. 2011; Menary et al. 2013; Shaw et al. 2006) as well as adult samples (Choi et al. 2008; Haier et al. 2004; Narr et al. 2007). Furthermore, the present posterior callosal association might reflect especially the maturation of the “parietal functions” within parieto-frontal integration theory of intelligence (PFIT; Jung and Haier 2007), and be associated in particular with early processing and integration of sensory information. A “non-functional” interpretation is also conceivable; any callosal variability is likely to reflect differences in the developing cortical architecture but might not itself contribute to the level of performance (Strauss et al. 1994). That is, the corpus callosum would serve as a “marker” for an “intelligent brain”.

However, the above interpretation deserves further clarification. When introducing chronological age into the statistical model, the suggested associations between callosal thickness and raw intelligence scores disappeared. In the studied period of life, between 6 and 22 years, it is known that intelligence test performance when expressed as raw scores (Tamnes et al. 2010, see also Fig. 1) and callosal thickness (e.g., Giedd et al. 1996; Luders et al. 2010b; Westerhausen et al. 2016) are characterized by a continuous, monotonous age-related increase. Consequently, the temporal co-occurrence of these two developmental trends alone could drive the present structure–function association without the two variables necessarily being interrelated (e.g., Salthouse 2011, for a more general discussion). Thus, to examine the interdependence beyond the co-occurring age-related covariance, we introduced chronological age to the design, which yielded the effect of Test Score for both p-RS and v-RS non-significant. This “null” finding by itself does not unequivocally exclude the existence of a true functional association between intelligence measures and callosal thickness. Assuming, for example, that callosal maturation would be the only determinant of intelligence test performance, both variables would exhibit a strong covariation of rates of change during development, a correction for age would likely leave the two variables uncorrelated. However, the also non-significant interaction of Test Score and Age further indicates that the lack of covariance of Test Score and callosal thickness is invariant across all ages studied (Hofer et al. 2006; Salthouse 2011). Thus, the lack of a significant effect of Test Sore together with the non-significant interaction of Test Score and Age, indicate that for no chronological age a significant structure–function association was found. Given sufficient test power to exclude effects larger than 2–3% explained variance, the present findings render any substantial functional relevance of macrostructural callosal variability for intelligence test performance unlikely, within the studied age range.

The non-relevance of individual differences in callosal thickness for intelligence test performance is, on one hand, not surprising as it is in line with studies on patients with partial or complete callosotomy (e.g., Mamelak et al. 1993; Oguni et al. 1991; Tanriverdi et al. 2009). Pre-post-surgery comparisons usually fail to find a substantial decrease in IQ scores as consequence of the callosal transsection. For example, Mamelak et al. (1993) examined 15 epilepsy patients aged between 9 and 31 years (i.e., also covering the age range examined here) and did neither find substantial change in verbal (mean change in v-IQ: 0.38, SD = 3.52; from Table 3, p. 692, of Mamelak et al. 1993) nor in performance IQ (mean change in p-IQ: 1.23, SD = 3.56). Nevertheless, it has to be kept in mind that findings in callosotomy patients are not fully representative, as the patients’ pre-surgical brain anatomy certainly deviates from a healthy brain including plastic adaption processes to the epilepsy-related changes (Campbell et al. 1981). On the other hand, the present findings are partly at odds with a series of earlier studies demonstrating that inter-individual difference in callosal architecture are predictive of functional inter-hemispheric integration in tasks assessing cognitive-control functions (Davis and Cabeza 2015; Huster et al. 2011; Kompus et al. 2011; Schulte et al. 2006). However, a critical component of these tasks was that speeded choice reactions were required, demanding a rapid exchange of information between the hemispheres. The performance in the four tasks which constitute the WASI test, on the other hand, is not relying on fast stimulus–response mapping in a similar manner. Also the association between general white-matter tract fractional anisotropy and intelligence measures has been previously shown to be mediated by general information-processing speed (Penke et al. 2012). Thus, it is conceivable that in the present study no structure–function association was found since time-critical communication between the hemispheres was not required to the same extend in the WASI tests.

The present findings also need to be discussed in relation to previous studies correlating deviation IQ scores with measures of callosal anatomy in comparable developmental samples. Irrespective of whether interpreting the outcome of the analyses without (positive association) and with Age covariates (no association), the present findings do not align with a series of previous developmental studies reporting negative associations (Allin et al. 2007; Ganjavi et al. 2011; Hutchinson et al. 2009; Luders et al. 2011). As also argued in the introduction, deviation IQ scores reflect the relative position of an individual within a specific norm age group and no longer the absolute level of performance (Angoff 1984). This is, for example, reflected in the present sample as raw scores in the prediction of IQ scores explains at maximum 42% of the variance (see “Method” section). Furthermore, the conversion from raw measure to IQ score also removes overall differences in performance level between individuals that fall into different age groups (different conversion table are used) and set to the mean to the norm distribution’s mean (Wechsler 1999). The converted IQ scores will consequently appear stable (Burgaleta et al. 2014) throughout childhood and adolescence while the absolute level of performance naturally rises continuously in this life period (see Fig. 1). At the same time, however, none of the previous studies converted (or adjusted to norm-data) callosal morphology measures. Rather, two of the above developmental studies (Allin et al. 2007; Luders et al. 2011) related deviation IQ scores to raw measures of midsagittal callosal area and regional callosal thickness, respectively, both without correcting for chronological age. Considering the properties of deviation IQ the reported findings can be reformulated to: the participants’ relative position in performance with respect to their age-specific norm group was negatively associated with the participants’ absolute measures of corpus-callosum size/thickness. In the present study, a positive correlation between absolute level of performance and absolute callosal thickness was detected (when not considering Age in the analysis). Of note, the attempt to replicate the negative associations of Luders et al. (2011) and Allin et al. (2007) using deviation IQ scores did not yield any significant associations in the present sample (see Supplementary Fig. 2). Two other studies (Ganjavi et al. 2011; Hutchinson et al. 2009) introduced chronological age as covariate into the statistical design while at the same time keeping the already age-adjusted deviation IQ score. While the covariate will remove age-related variance also from the callosal measures by implicitly treating these as deviation from the predicted mean for the respective age, it does not express the data relative to an external norm distribution as it is done for deviation IQ score, again making the interpretation difficult. Considering Age as covariate and in interaction with Test Scores in the present study, did not reveal any significant association between test performance and absolute thickness.

Finally, previous studies also report sex differences in the association of callosal and deviation IQ measures (Dunst et al. 2014; Luders et al. 2007; Tang et al. 2010), whereby in developmental samples a more negative associations in male compared to female subsamples has been reported (Ganjavi et al. 2011; Luders et al. 2011). In the present study, we find a Test score by Sex interaction in the genu of the corpus callosum for v-RS and p-RS in analysis step 1, which was in both cases driven by a positive association in females while no significant association was found in male participants (see Fig. 4). As axons located in the genu interconnect prefrontal cortices (Benedictis et al. 2016; Schmahmann and Pandya 2006) the findings might relate to the frontal components of intelligence (Jung and Haier 2007), whereby selectively female participants seem to benefit from a stronger structural connectivity via the corpus callossum. However, this dissociation only holds for “absolute” callosal measures; once TIV was accounted for the interaction did not survive FDR correction.

In summary, intellectual abilities are supported by a large-scale bihemispheric network (Deary 2012; Jung and Haier 2007; Shaw 2007) suggesting functional relevance of inter-hemispheric coordination (Banich 2003; Davis and Cabeza 2015). The present study was able to confirm that a general structure–function correlation exists during development but only as long as the participants’ age was not considered. Thus, we did not find any association that cannot be explained by a temporal co-occurrence of overall developmental trends in intellectual development and structural callosal increase. However, we here examined the macrostructural development using regional thickness measures. It remains for future studies to determine whether these findings are confirmed when indices of microstructural differences, such as fractional anisotropy, are assessed. Future studies also have to establish, whether the present finding utilizing RS instead of deviation-IQ extend into adult age, as the structural development of the corpus callosum continuous at least to the end of the third life decade (e.g., Pujol et al. 1993).