Sex chromosome aneuploidy alters the relationship between neuroanatomy and cognition

Abstract

Sex chromosome aneuploidy (SCA) increases the risk for cognitive deficits, and confers changes in regional cortical thickness (CT) and surface area (SA). Neuroanatomical correlates of inter-individual variation in cognitive ability have been described in health, but are not well-characterized in SCA. Here, we modeled relationships between general cognitive ability (estimated using full-scale IQ [FSIQ] from Wechsler scales) and regional estimates of SA and CT (from structural MRI scans) in both aneuploid (28 XXX, 55 XXY, 22 XYY, 19 XXYY) and typically-developing euploid (79 XX, 85 XY) individuals. Results indicated widespread decoupling of normative anatomical–cognitive relationships in SCA: we found five regions where SCA significantly altered SA–FSIQ relationships, and five regions where SCA significantly altered CT–FSIQ relationships. The majority of areas were characterized by the presence of positive anatomy-IQ relationships in health, but no or slightly negative anatomy-IQ relationships in SCA. Disrupted anatomical–cognitive relationships generalized from the full cohort to karyotypically defined subcohorts (i.e., XX-XXX; XY-XYY; XY-XXY), demonstrating continuity across multiple supernumerary SCA conditions. As the first direct evidence of altered regional neuroanatomical–cognitive relationships in supernumerary SCA, our findings shed light on potential genetic and structural correlates of the cognitive phenotype in SCA, and may have implications for other neurogenetic disorders.

1 |. INTRODUCTION

The carriage of abnormal numbers of sex chromosomes (i.e., sex chromosome aneuploidy, SCA) is correlated with heightened risk for cognitive deficits, as well as changes in cortical thickness (CT) and surface area (SA) (Lepage et al., 2014; Raznahan et al., 2016). Evidence from structural MRI (sMRI) studies in typically developing groups suggests that variation in CT and SA is associated with variation in general cognitive ability (Schmitt, Neale, et al., 2019; Schmitt, Raznahan, et al., 2019; Shaw et al., 2006; Tadayon, Pascual-Leone, & Santarnecchi, 2019). Studying this relationship in SCA would help clarify the relevance of neuroanatomical changes for understanding SCA-associated cognitive deficits. It could also provide insight into the broader question of biological pathways to altered cognition in other neurogenetic disorders associated with both neuroanatomical and cognitive changes (i.e., 16p11.2 variation: Hippolyte et al., 2016; 22q11.2 variation: Lin et al., 2020): Although cortical structure and cognition both have strong, and perhaps even shared, underlying genetic components (Peper, Brouwer, Boomsma, Kahn, & Pol, 2007; Polderman et al., 2015; Schmitt, Raznahan, et al., 2019; Sniekers et al., 2017), we still lack a complete understanding of their relationship in these genetic conditions.

The inherent individual differences in cognition (Deary, Penke, & Johnson, 2010), and known implications those differences have for social outcomes (i.e., educational attainment, job performance, and mortality; Deary, Weiss, & Batty, 2010; Lynn, Fuerst, & Kirkegaard, 2018), have motivated sustained study of the relationship between inter-individual structural and cognitive variability in typically-developing populations (Kanai & Rees, 2011). Results are some-what mixed, but generally support a distributed, directionally-positive model of cognition and structure in health, with increases in cognitive abilities correlated with increases in both total brain volume (TBV) (Cox, Ritchie, Fawns-Ritchie, Tucker-Drob, & Deary, 2019; Mcdaniel, 2005; Pietschnig, Penke, Wicherts, Zeiler, & Voracek, 2015) and regional gray matter volume (GMV) (Basten, Hilger, & Fiebach, 2015; Jung & Haier, 2007). Recent studies have also related variation in cognition to variation in the neurobiologically dissociable subcomponents of GMV, CT and SA (Panizzon et al., 2009; Rakic, 1995). Widely-distributed positive relationships between cognitive abilities and regional CT have been noted, with convergent findings in the prefrontal cortex, inferior temporal gyri, temporal poles, precuneus, and occipital lobe (Bajaj et al., 2018; Choi et al., 2008; Karama et al., 2009; Menary et al., 2013; Narr et al., 2007; Schmitt, Raznahan, et al., 2019). Fewer designs have incorporated SA, but significant cognition–SA correlations have been identified in dispersed subregions of the prefrontal cortex, middle frontal lobe, inferior temporal gyrus, lateral superior parietal lobe, and precuneus (Román et al., 2014; Schmitt, Neale, et al., 2019; Tadayon et al., 2019; Walhovd et al., 2016). Importantly, regions of significant cognition–structure relationships are often distinct for SA and CT within cohorts (Román et al., 2014; Tadayon et al., 2019; Walhovd et al., 2016).

SCA is associated with both anatomical and cognitive changes, which have been well-characterized by independent research groups. Compared to karyotypically normal cohorts, patients with supernumerary SCA generally show moderate decreases in overall cognitive functioning, with particular impairments in verbal IQ and language skills (Bishop & Scerif, 2011; Lee et al., 2012; Leggett, Jacobs, Nation, Scerif, & Bishop, 2010; Visootsak, Rosner, Dykens, Tartaglia, & Graham, 2007). Anatomically, although supernumerary X- and Y-chromosomes confer divergent changes in TBV (+ X = ↓ TBV; + Y = ↑ TBV [Bryant et al., 2011; Raznahan et al., 2016]), extensive prior evidence suggests that they have convergent effects on both regional cortical anatomy (i.e., + X | +Y = ↑ CT in the prefrontal cortex [Lepage et al., 2014; Raznahan et al., 2016]) and subcortical structures (Mankiw et al., 2017; Nadig et al., 2018; Reardon et al., 2016). SCA also appears to have dissociable effects on SA and CT (Raznahan et al., 2016). Little is known, however, about the relationship between anatomical variation and cognitive outcomes in SCA. The few studies on this topic to date have yielded varied results, and employed analysis of cortical GMV (rather than separately considering SA and CT) in modestly-sized cohorts (Brown et al., 2004; Bryant et al., 2011; DeLisi et al., 2005; Itti et al., 2006; Patwardhan, Eliez, Bender, Linden, & Reiss, 2000; Seiler et al., 2018; Skakkebæk et al., 2014; Warwick et al., 1999). Here, using a unique dataset of euploid (79 XX, 85 XY) and aneuploid (28 XXX, 55 XXY, 22 XYY, 19 XXYY) participants, we sought to extend current understandings by directly modeling the association between inter-individual variability in neuroanatomy and inter-individual variability in cognition in both health and SCA. We calculated regional estimates of CT and SA from surface-based analysis of sMRI scans, and used Weschler-derived full-scale IQ (FSIQ) as a measure of general cognitive ability. We pursued stepwise analyses which (a) determined regions of significant SA–FSIQ and CT–FSIQ relationships in our healthy control cohort, (b) determined regions showing significant disruptions of those normative SA–FSIQ and CT–FSIQ relationships in SCA, (c) compared the regions with disrupted SA–FSIQ or CT–FSIQ relationships in SCA to those showing significant SA or CT changes in SCA, and (d) verified the stability of our findings across multiple aneuploid subgroups.

2 |. METHODS

2.1 |. Editorial policies and ethical considerations

This study was approved by the National Institutes of Health (NIH) Institutional Review Board, and informed consent was obtained from all participants.

2.2 |. Participants

We included 288 participants of varying karyotypes. Participant characteristics are described in Figure 1a. Individuals with SCA were recruited through parent support groups and the NIH website. Supernumerary X-/Y-chromosome carriage was confirmed with karyotype tests. Typically-developing euploid controls were sampled from the NIH Longitudinal Structural MRI Study of Human Brain Development (Giedd et al., 2015). Exclusionary criteria for SCA participants included a history of head injury, gross brain abnormalities, and mosaicism, with nonmosaicism confirmed by visualization of 50 metaphase spreads in peripheral blood. Exclusionary criteria for euploid controls included a history of mental illness, use of psychiatric medication, diagnosis of a nervous system disorder, and enrollment in special education services. Of note, this cohort has previously been used to examine sex and SCA effects on cortical and subcortical structures (Fish et al., 2017; Lin et al., 2015; Mankiw et al., 2017; Nadig et al., 2018; Raznahan et al., 2016; Xenophontos et al., 2019).

FIGURE 1.

Sample characteristics and full-scale IQ (FSIQ) phenotype boxplots. (a) Detailed participant characteristics, arranged by karyotypic subgroups. (b) Boxplots of FSIQ scores, arranged by group (control vs. sex chromosome aneuploidy [SCA]), and karyotypic subgroups shown as colored points. Mean FSIQ score for SCA participants was significantly lower than the mean FSIQ score within the control cohort (t [250] = 12.4, p < .001)

2.3 |. Image acquisition and processing

All sMRI scans were T1-weighted, and were obtained on the same 1.5 Tesla General Electric SIGNA scanner with contiguous 1.5 mm axial slices, using a 3D spoiled gradient-recalled echo sequence with the following acquisition parameters: echo time, 5 ms; repetition time, 24 ms; flip angle, 45°; acquisition matrix, 256 × 192; number of excitations, 1; field of view, 24 cm.

Native scans were submitted to the CIVET 1.1.10 pipeline for automated morphometric analysis (Ad-Dab’bagh et al., 2006). After initial correction of images for nonuniformities in radiofrequency intensity (Collins, Neelin, Peters, & Evans, 1994; Sled, Zijdenbos, & Evans, 1998), this analysis used a validated neural net approach to voxel classification to estimate gray and white matter volumes (Cocosco, Zijdenbos, & Evans, 2003; Zijdenbos, Forghani, & Evans, 2002). Next, as described previously (Raznahan et al., 2016), images were fitted with two deformable mesh models to identify the inner and outer surfaces of cortical gray matter, and those surfaces were used to calculate CT and SA at 81,924 vertices across the cortical sheet. Finally, the vertex-level data were downsampled using a subparcellation of the Desikan-Killiany atlas (Romero-Garcia, Atienza, Clemmensen, & Cantero, 2012), generating estimates of mean CT and SA at 308 roughly equally-sized (~5 cm²), spatially-contiguous regions. All included scans passed rigorous quality assessment procedures, including verifying absence of visible artifacts in raw scans and in cortical surfaces extracted from CIVET.

2.4 |. General cognitive ability

FSIQ was estimated for all participants using an age-appropriate Weschler test. Most participants (n = 201) received the Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999). Other scales included the WAIS-III, WAIS-R, WISC-III, WISC-R, WPPSI, WPPSI-III, and WPPSI-R.

2.5 |. Experimental design and statistical analyses

2.5.1 |. Primary analyses

Primary analyses separately modeled SA and CT for each region as follows (Model 1):

$$\begin{gathered} {\text{Anatomy}}_{i\hspace{0.17em}\text{residualized}}=\text{Intercept}+{\beta }_1{\text{FSIQ}}_{\text{residualized}}+{\beta }_2\text{Group}+{\beta }_3\text{Group} \\ *{\text{FSIQ}}_{\text{residualized}}+\text{error} \end{gathered}$$

where Anatomy_{i residualized} is mean SA or CT across vertices in region i, corrected for the effects of chromosomal subgroup (i.e., XX, XY, XXX, XXY, XYY, or XXYY) and age, group is a binary categorical variable denoting either euploid control or SCA group status (reference category = control group), FSIQ_residualized denotes FSIQ measures corrected for the effects of karyotypic subgroup, and group*FSIQ_residualized is the interaction effect between group status and residualized FSIQ. Correcting for karyotypic subgroup ensured any inter-subgroup effects did not obscure analysis of inter-individual effects; correcting for age is common practice in cross-sectional sMRI neuroanatomy studies, and harmonized anatomical measures with FSIQ, which is inherently age-normed. Left and right hemispheres were modeled separately, so estimation of this model within each cortical region created vectors of 152 (left hemisphere) or 156 (right hemisphere) test statistics associated with each of the β₁, β₂, and β₃ coefficients.

To verify that FSIQ–SA and FSIQ–CT relationships were present in our euploid control cohort, and that they were in line with previous findings in healthy controls, p-values associated with the β₂ coefficients from this model were corrected for multiple comparisons using False Discovery Rate (FDR) correction (q = .05), and t-statistics for surviving β₂ coefficients were projected onto a reference cortical surface for visualization. Surviving regions were those where there was a significant relationship between FSIQ and SA or FSIQ and CT in euploid controls.

To test for regions where there was a significant interaction effect of SCA group status on CT–cognition relationships or SA–cognition relationships, p-values associated with the β₃ coefficients from this model were corrected for multiple comparisons using FDR (q = .05), and t-statistics for surviving β₃ coefficients were projected onto the reference cortical surface. Surviving areas were those where SCA status had a significant effect on the relationship between FSIQ and CT or FSIQ and SA.

2.5.2 |. Effect of SCA on anatomy

Regions showing significant interaction effects of SCA status on FSIQ–CT or FSIQ–SA relationships were then compared to those where there were significant between-group differences in anatomical measures only (i.e., those where SCA status had a significant effect on SA or CT). Relationships between SCA status and CT and SA were modeled at each region as follows (Model 2):

$${\text{Anatomy}}_i=\text{Intercept}+{\beta }_1\text{Group}+{\beta }_2\text{Age}+{\beta }_3\text{Sex}+\text{error}.$$

where anatomy_i is mean SA or CT across vertices in region i, group is a binary categorical variable denoting either euploid control or SCA group status (reference category = control group), age is a continuous variable identifying subject age, and sex is a binary categorical variable (XY | XXY | XYY | XXYY = male, XX | XXX = female; reference category = female). Again, left and right hemispheres were modeled separately.

For this model, p-values associated with the β₁ coefficients were corrected for multiple comparisons using FDR (q = .05), and t-statistics for surviving β₁ coefficients were projected onto the reference cortical surface. Surviving areas were those where there were significant effects of SCA status on either CT or SA.

To examine how regions where SCA affected anatomy–cognition relationships were related to those showing anatomical changes in SCA, we compared the spatial layout of the regions with significant SCA on CT–FSIQ and SCA on SA–FSIQ interaction effects from the primary analysis to areas showing significant SCA effects on CT and SA from the secondary analysis. To do this, for both SA and CT, we counted the number of areas showing significant interaction effects in Model 1 that coincided with areas showing significant main effects of group in Model 2. We compared the coinciding counts (one for SA, one for CT) to respective null distributions of SA or CT counts calculated by repeating the process with 10,000 random permutations of the interaction map from Model 1 checked against the true map of regions showing significant main effects in Model 2.

2.5.3 |. Sensitivity analyses

Finally, we ran two sensitivity analyses to ensure the integrity of our primary analytic framework.

1. Consideration of total brain volume.

Previous work has shown that TBV is associated with FSIQ in health (Cox et al., 2019) and SCA (Warwick et al., 1999). To ensure that variation in TBV did not affect regional-level correlations, we included an age- and karyotypic subgroup-corrected measure of TBV as a coefficient in Model 1, and correlated (Pearson’s r) t statistics associated with the β₂ and β₃ coefficients to those from the original output of the model. For this analysis, hemispheres were combined, generating four r values in total: r_{β2 CT}, r_{β2 SA}, r_{β3 CT}, and r_{β3 SA}.

2. Replication in SCA subgroups.

The above approaches compared one euploid control group to one SCA group (i.e., collapsed across aneuploidy subgroups), which improved statistical power for delineating group differences, and was supported by prior evidence showing convergence of X- and Y-chromosome effects on regional anatomy (Lepage et al., 2014; Raznahan et al., 2016). However, this raises the possibility that observed results were driven by the heterogeneity of the SCA sample. To verify that the results for the combined SCA group generalized to individual SCA subgroups, we re-ran the primary interaction analysis in three subcohorts: one with only XX and XXX participants (28 XX, mean age = 11.93; 28 XXX, mean age = 12.33), one with only XY and XXY participants (55 XY, mean age = 12.56; 55 XXY, mean age = 12.95), and one with only XY and XYY participants (22 XY, mean age = 12.90; 22 XYY, mean age = 13.22). Subcohorts were generated by including all aneuploid subjects in the particular SCA subgroup, and then randomly sampling an equal number of gonadally- and age-matched euploid controls from the full cohort. We did not include an XXYY-specific subcohort because of the small sample size (n = 19 XXYY); the XYY subcohort (n = 22 XYY) was included to give a comprehensive view of generalizability within all trisomy cohorts.

As in the primary analysis in the full cohort, regional CT and SA were modeled using Model 1, which generated test statistics associated with the β₁Group, β₂FSIQ_residualized, and β₃Interaction coefficients. Because we were interested in the generalizability of the interaction effects of SCA on FSIQ–CT and FSIQ–SA relationships, we focused on the β₃ coefficients.

To confirm coherence of interaction effects between the full SCA group and each SCA subgroup, we correlated (Pearson’s r) uncorrected t-value vectors associated with the β₃ coefficients in each SCA subcohort with the uncorrected t-value vectors associated with the β₃ coefficients in the full SCA cohort. For this analysis, right and left hemispheres were combined, generating 6 r values in total: r_{XXX-Full group CT}, r_{XXX-Full group SA}, r_{XXY-Full group CT}, r_{XXY-Full group SA}, r_{XYY-Full group CT}, and r_{XYY-Full group SA}. Observed r values were compared with null distributions of r values generated by repeating the analyses with 1,000 t-statistic vectors derived from permutation of aneuploidy status within each subgroup.

Additionally, for visualization purposes, we separately modeled left and right hemispheres in the subgroups using Model 1, and projected z-scores associated with unthresholded t-statistics for the β₃ coefficients onto the reference cortical surface.

All statistical models were generated using R 3.6.0 (R Core Team, 2013).

3 |. RESULTS

3.1 |. General cognitive ability

FSIQ significantly differed between the SCA and control groups (t[250] = 12.4, p < .001), with SCA scores (mean = 93.1, SD = 15.8) being, on average 22.4 points lower than those in controls (mean = 115.5, SD = 14.3) (Figure 1b).

3.2 |. Anatomical–cognitive relationships in healthy controls

We first verified the presence of relationships between our anatomical measures of interest, CT and SA, and FSIQ scores in our euploid control cohort. We found many regions with significant, positive, CT–FSIQ and SA–FSIQ relationships. However, there was little spatial overlap of regions between the two anatomical measures: only the left middle frontal gyrus, left inferior temporal gyrus, and left lateral occipital lobe showed positive relationships between FSIQ and both SA and CT. Outside of those regions, significant CT–FSIQ relationships were present bilaterally in the temporal and frontal poles, and unilaterally in several other, well-distributed cortical areas (see Figure S1A). Significant SA–FSIQ relationships were present bilaterally within the superior frontal gyri, cuneus, and entorhinal cortices, and unilaterally in several different, but also well-distributed, cortical areas (see Figure S1B). We found no significant negative relationships between FSIQ and either anatomical measure in our control cohort.

3.3 |. Effects of SCA on anatomical–cognition relationships

We next examined how SCA status affected anatomical–cognition relationships that were present in health. SCA group status had a significant impact on four regional CT–FSIQ relationships in the right hemisphere (laterally in the ventral superior parietal lobe and middle temporal gyrus, and medially in the precuneus and cuneus), and in the temporal pole in the left hemisphere (Figure 2a). SCA status also had a significant effect on five SA–FSIQ relationships, but in different regions: in one area in the right lateral superior parietal lobe, in two areas in the right lateral occipital lobe, in one area in the right lingual gyrus, and one area in the left dorsal postcentral gyrus (Figure 2b).

FIGURE 2.

Interaction effect of sex chromosome aneuploidy on regional anatomical–cognitive relationships. (a) Cortical regions where sex chromosome aneuploidy (SCA) status had a significant effect on relationships between cortical thickness (CT) and full-scale IQ (FSIQ), after controlling for age and karyotypic subgroups. (b) Cortical regions where SCA status had a significant effect on relationships between surface area (SA) and FSIQ, after controlling for age and karyotypic subgroups. Positive interaction effects are in red; negative interaction effects are in blue. (c) Representative region (precuneus, pCUN), showing within-group trends for FSIQ–CT relationships in the control and SCA groups. (d) Representative region (dorsal postcentral gyrus, dPG), showing within-group trends for FSIQ–SA relationships in the control and SCA groups. P-values refer to the interactive effects of group and IQ on anatomy for each region. With the exception of the superior parietal lobule region (sPL), all regions with significant interaction effects showed this trend of positive FSIQ–anatomical relationships in controls, but no or slightly negative FSIQ–anatomical relationships in SCA. CUN, cuneus; clOL, caudal lateral occipital lobe; LG, lingual gyrus; mTG, medial temporal gyrus; TP, temporal pole; rlOL, rostral lateral occipital lobe; vsPL, ventral superior parietal lobe

Although there was no spatial overlap between areas with significant CT effects and areas with significant SA effects, strikingly, with the exception of the superior parietal lobe SA–FSIQ relationship, all interaction effects were driven by the same pattern: the presence of a positive relationship between FSIQ variation and regional variation in the anatomical measure in controls, but either no relationship or a slightly negative relationship in the same area in SCA (Figure 2c,d). The superior parietal region alone was characterized by the opposite pattern—no FSIQ–SA relationship was present in controls, but there was a positive SA–FSIQ relationship in the SCA group.

3.4 |. Effect of TBV on regional models in health and SCA

To ensure that TBV did not affect regional FSIQ–anatomy correlations, we added a measure of TBV as a term to Model 1, and then correlated (Pearson’s r) t statistics associated with the β₂ and β₃ coefficients from the original model with those from the model incorporating TBV. Correlations were highly significant for the β₂ coefficients (r_{β2 CT} = .97; p < 2.2e–16; r_{β2 SA} = .97; p < 2.2e–16) and the β₃ coefficients (r_{β3 CT} = .99; p < 2.2e–16; r_{β3 SA} = .99; p < 2.2e–16). These results indicate that TBV did not meaningfully alter regional CT–FSIQ or SA–FSIQ relationships within our control cohort (β₂ correlations) nor meaningfully alter the interaction effects of SCA on regional CT–FSIQ or SA–FSIQ relationships (β₃ correlations). Thus, we did not include a measure of TBV in any of our final models.

3.5 |. Comparison to areas with CT or SA sensitive to SCA

We next compared the 10 areas showing significant interaction effects of SCA status on FSIQ–CT and FSIQ–SA relationships to regions where CT or SA was significantly affected in SCA. All five of the CT–FSIQ interaction regions of interest fell within regions where SCA had a significant effect on CT. Permutation testing (see Section 2) revealed this correspondence was significantly greater than null expectations (p_Perm = 0.002). However, only two of the five SA–FSIQ interaction regions of interest fell within regions where SCA had a significant effect on SA. Permutation testing indicated the overlap was likely due to chance for SA (p_Perm = 0.800), suggesting SCA effects on CT were more salient for the disrupted CT–FSIQ relationships than were SCA effects on SA for disrupted SA–FSIQ relationships.

3.6 |. Generalizability to SCA subgroups

Finally, to ensure generalizability of results found using the combined SCA group to individual SCA chromosomal subgroups, we re-examined the effect of SCA status on CT–FSIQ and SA–FSIQ relationships in three subcohorts: one including only XX and XXX participants, one including only XY and XXY participants, and one including only XY and XYY participants (Figure 3a,b). Pearson correlations indicated a statistically-significant spatial coherence between combined SCA sample effects and XXX, XXY, and XYY effects on FSIQ–CT (r_{XXX-Full group CT = 49}, p < 2.2e–16; r_{XXY-Full group CT = .57}, p < 2.2e–16; r_{XYY-Full group CT} = .37, p = 1.8e–11; Figure 3c) and FSIQ–SA (r_{XXX-Full group SA} = .53, p < 2.2e–16; r_{XXY-Full group SA} = .65, p < 2.2e–16; r_{XYY-Full group SA} = .32, p = 1.6e–8; Figure 3d) relationships. Permutation testing (see Section 2) confirmed these close relationships were due to coherence between full and subcohort SCA effects (p_Perm ≤ 0.01 in all cases; see Figure 3c,d), indicating that observed aberrant SCA effects on cognitive–anatomical relationships held for both the combined SCA group and for individual SCA chromosomal subcohorts.

FIGURE 3.

Replication of interaction effects from full sample to three karyotypically-defined subcohorts. (a) Unthresholded maps showing the interaction effect of sex chromosome aneuploidy (SCA) status on regional relationships between cortical thickness (CT) and full-scale IQ (FSIQ) within the full sample, a cohort consisting of XX and XXX participants, a cohort consisting of XY and XXY participants, and a cohort consisting of XY and XYY participants. (b) Unthresholded maps showing the interaction effect of SCA status on regional relationships between surface area (SA) and FSIQ within the full sample and same XX-XXX, XY-XXY, and XY-XYY cohorts. (c) Scatterplots illustrating the close coherence between interaction effects of SCA status on CT–FSIQ relationships between the full cohort and the XX-XXX, XY-XXY, and XY-XYY cohorts; and permutation testing distributions (with 1,000 null r values) demonstrating that observed correlations are significantly greater than null expectations. (d) Scatterplots illustrating close coherence between interaction effects of SCA status on SA–FSIQ relationships between the full cohort and XX-XXX, XY-XXY, and XY-XYY cohorts; as well as permutation testing distributions showing observed correlations are significantly greater than null expectations

4 |. DISCUSSION

Using a unique cohort of youth with varying sex chromosome dosage, we replicate and extend current conceptions of anatomical–cognition relationships in both typical development and disease.

4.1 |. Anatomy–cognition relationships in health

In our euploid controls, we found that FSIQ was positively correlated with the two primary biologically distinct (Panizzon et al., 2009; Rakic, 1995; Raznahan et al., 2011) anatomical dimensions of the cortical sheet, SA and CT (Figure S1). These positive associations were present in widely-dispersed subregions of all four cortical lobes, and we found no significant negative FSIQ–CT or FSIQ–SA relationships. The observed positive correlations overlap with those reported in prior studies (CT: [Bajaj et al., 2018; Karama et al., 2009; Menary et al., 2013; Narr et al., 2007; Schmitt, Raznahan, et al., 2019; Shaw et al., 2006], SA: [Schmitt, Neale, et al., 2019; Tadayon et al., 2019; Walhovd et al., 2016]). Further, our findings are broadly consistent with a prominent structure–function model of cognition, the Parietal–Frontal Integration Theory (P-FIT), which implicates gray matter within occipito-temporal sensory cortices, as well as in higher-order frontal and parietal association areas, in improved cognitive functioning (Jung & Haier, 2007). Of note, however, regions of strongest correlation with FSIQ variation were largely distinct for SA (superio-medial prefrontal, lateral temporal and superior parietal cortex) and CT (temporo-parietal junction, temporal pole, dorsolateral prefrontal cortex), demonstrating the value of considering these two cortical dimensions separately.

4.2 |. Anatomy–cognition relationships in SCA

Through comparison of euploid controls and SCA patients, we provide the first direct evidence of alterations in regional anatomical–cognitive relationships in supernumerary X-/Y-conditions (Figure 2). Our study design builds on prior work in supernumerary SCA, which did not find such alterations (Bryant et al., 2011; DeLisi et al., 2005; Patwardhan et al., 2000; Skakkebæk et al., 2014), by including a significantly expanded sample size encompassing multiple SCA groups, and querying cortical anatomy using surface-based methods capable of dissociating between CT and SA. Through a combination of these approaches, we detected widespread disruption of normative regional anatomical–cognitive relationships in our SCA participants, including five regions where SCA significantly altered CT–FSIQ relationships (which were dispersed throughout the occipital, parietal, and temporal lobes), and five regions where SCA significantly altered SA–FSIQ relationships (which lay within the occipital and parietal lobes). The vast majority of these regions were characterized by apparent decoupling of normative anatomical–cognitive relationships in SCA (i.e., there was a positive anatomical–FSIQ relationship in health, but no discernable anatomical–FSIQ relationship in SCA). Echoing our observations for anatomy–FSIQ relationships in health, the regional disruptions of anatomy–FSIQ relationships in SCA were largely nonoverlapping for CT and SA. Moreover, all five FSIQ–CT regions of interest overlapped with regions showing significant CT changes in SCA, whereas this did not apply for SA. This striking observation suggests that the mechanisms through which SCA alters CT in the superior parietal lobe, temporal pole, cuneus, and precuneus may be particularly important for the disruption of normative structure–function relationships in groups with supernumerary X- and Y-chromosomes. Finally, we demonstrated through sensitivity analyses that the maps of disrupted FSIQ–anatomy relationships observed across all SCAs combined were significantly correlated with those observed when three SCA subgroups were considered in isolation (Figure 3). This finding attests to the reproducibility of X-chromosome dosage effects on the brain across both males and females (Mankiw et al., 2017; Nadig et al., 2018; Raznahan et al., 2016; Reardon et al., 2016), and further strengthens the growing evidence base for a general convergence between the effects of supernumerary X- and Y-chromosomes on human brain organization (Mankiw et al., 2017; Nadig et al., 2018; Raznahan et al., 2016; Reardon et al., 2016).

4.3 |. Implications of disturbed anatomical–cognitive relationships in SCA

The patterns of disrupted anatomical–cognitive relationships we observed in supernumerary SCA help to refine hypotheses regarding the biological bases of altered brain function, as well as altered structure–function relationships, in groups carrying supernumerary sex chromosomes, and potentially in neurogenetic disorders more generally.

By defining specific cortical regions of disrupted anatomy–FSIQ interrelationships in SCA, our findings nominate a subset of brain regions that may be particularly relevant for the decrements in general cognitive functioning that have been reported in SCA (Hong & Reiss, 2014; Joseph et al., 2018; van Rijn, 2019). In particular, it is notable that we do not observe disrupted anatomy–FSIQ interrelationships within prefrontal cortex regions that have previously been linked to the highest levels of cognitive integration, evaluation, and response selection (Barbey, Colom, & Grafman, 2013; Jung & Haier, 2007; Miller & Cohen, 2001). Rather, most foci of disrupted anatomy–FSIQ interrelationships in SCA were found within subregions of temporal, parietal, and occipital lobes that are known to play important roles in the reception of primary visual (cuneus/lateral occipital lobe: Kosslyn, 1999) and somatosensory (postcentral gyrus: Kaas, Nelson, Sur, Lin, & Merzenich, 1979; Sanchez-Panchuelo, Francis, Bowtell, & Schluppeck, 2010) information, integration of visuo-spatial information and memory retrieval (precuneus: Cavanna & Trimble, 2006; lingual gyrus: Bogousslavsky, Miklossy, Deruaz, Assal, & Regli, 1987; Mechelli, Humphreys, Mayall, Olson, & Price, 2000; Sulpizio, Committeri, Lambrey, Berthoz, & Galati, 2013) and multimodal integration and processing (anterior middle temporal gyrus/temporal pole: Wong & Gallate, 2012; superior parietal lobe: Hawkins, Sayegh, Yan, Crawford, & Sergio, 2012; Heim et al., 2012; Wolpert, Goodbody, & Husain, 1998). Thus, our data raise the possibility that SCA impacts on general cognitive functioning may involve disruptions within the earlier receptive and initial integrative stages of cortical processing, rather than later stages (Jung & Haier, 2007).

Another key outcome from our study is that we do not observe cortical regions that show prominent correlations with FSIQ in SCA that are absent in euploid controls. In other words, SCA does not appear to be accompanied by a different network of anatomical correlations with FSIQ as compared to controls; rather, SCA is characterized by an absence of any discernible relationships in regions where there are correlations in controls. This observation could arise through a number of distinct biological phenomena at the systems-level. For example, the functional anatomy of FSIQ variation may be more individualized in SCA than in euploid controls, with a more idiosyncratic engagement of distributed brain systems in SCA. Alternatively, the same sets of functional networks could underpin FSIQ variation in SCA and controls, but the topography of these networks may be more variable across individuals with SCA than across individual controls. Through refinement of such distinct hypotheses, our study helps to direct future multimodal neuroimaging studies characterizing variation in the anatomical topography of functional features that predict varying cognitive impairment in SCA. However, it is also possible that disrupted structural–cognitive correlations in SCA are rooted in a loss of normative relationships between cognitive functions and cellular aspects of cortical organization. For example, increased dendritic arborization of pyramidal neurons has been associated with increased CT and higher FSIQ scores in healthy controls (Goriounova et al., 2018). Subtle changes in dendritic arborization could therefore alter both CT and FSIQ in SCA. Another potential mechanism involves white matter—the integrity of underlying white matter tracts may influence regional CT (Storsve, Fjell, Yendiki, & Walhovd, 2016; Tamnes et al., 2010), and has also been linked to cognitive abilities in health (Penke et al., 2012). Thus, disruption in white matter tracts, which has previously been noted in SCA (Blumenthal et al., 2013; Goddard, van Rijn, Rombouts, & Swaab, 2016), could have adverse effects on both cortical anatomy and cognitive functioning in the disorder.

A third implication of our findings relates to the nature of potential gene-dosage contributions to altered brain structure and function in SCA. Specifically, as already highlighted above, we found that the topography of altered anatomy–cognition relationships in SCA in the full mixed SCA cohort significantly correlated with topography from analyses in each of three karyotypically distinct subgroups: XXX, XXY and XYY. Because the effects adhered across three distinct gonadally-matched cohorts (i.e., one female, with XX and XXX participants; two male, one with XY and XXY participants, and one with XY and XYY participants), they likely reflect direct consequences of chromosome dosage variation in the brain, rather than mediation by the secondary (e.g., endocrine or metabolic) alterations that can also accompany SCA (Raznahan et al., 2016). Moreover, because we observe similar effects in both X- and Y-chromosome supernumeracy, the most likely causal genes are those that are shared by the X- and Y-chromosome (Raznahan et al., 2016, 2018). Previous work has proposed X-/Y-homologs outside the pseudoautosomal regions (“gametologs”; Bellott et al., 2014) as mediators of shared X and Y influences on neuroanatomic changes in SCA, because they have similar roles across both sex chromosomes, are closely tied to critical cellular functions (i.e., regulation of transcription and translation; Bellott et al., 2014) and appear to be unusually sensitive to changes in sex chromosome dosage (Raznahan et al., 2016, 2018; Xenophontos et al., 2019). Given that cortical structures and cognitive abilities appear to share genetic influences (Hulshoff Pol et al., 2006; Schmitt, Raznahan, et al., 2019), gametologs may also be good candidate drivers of the altered CT–FSIQ and SA–FSIQ relationships we observed in SCA.

Finally, our findings may be of relevance to other neurogenetic conditions outside of supernumerary SCA. Prior work in 45,X monosomy (Turner’s syndrome) has reported significant relationships between FSIQ scores and GMV in the postcentral gyrus (Brown et al., 2004), a region we found to have altered SA–FSIQ correlations in supernumerary SCA, and a study testing for differences between a Turner’s cohort and a healthy control cohort found clusters of altered relationships between GMV and a variety of cognitive domains in the precuneus and cuneus (Seiler et al., 2018), regions of aberrant CT–FSIQ correlations in our SCA sample. Thus, there appears to be some spatial correspondence for the effects of both sex chromosome loss and addition on regional anatomical–cognitive relationships. Additionally, a recent study found several distributed cortical regions that showed positive CT-processing speed relationships in health, but no CT-processing speed relationships in 22q11.2 deletion/duplication patients (Lin et al., 2020), which suggests some, albeit preliminary, coherence of the ways in which normative anatomical–cognitive relationships are disrupted across SCA and another neurogenetic condition. Thus, our current findings in supernumerary SCA add to a growing literature which highlights the power of genetically-defined conditions to help elucidate both phenotypic and biological understandings of altered behavioral/cognitive processes in disease. Lastly, the apparent alterations in normative structure–function relationships in neurogenetic disease illustrate a greater point of relevance for translational medicine: structure–function models that are defined in typically developing populations may not be applicable to clinical populations. This is important to consider given growing interest in the opportunities and challenges of precision medicine for genetic disorders.

4.4 |. Limitations and future directions

Our findings should be interpreted with consideration of several caveats. First, our analyses focused on the link between FSIQ and neuroanatomy. Although FSIQ is a widely-accepted proxy of general cognitive ability, there are subdimensions of cognitive impairment which have been described in SCA in the past (i.e., verbal learning deficits: Bishop & Scerif, 2011; Lee et al., 2012; Leggett et al., 2010; Visootsak et al., 2007) that also warrant separate comparison with neuroimaging metrics. Second, our typically-developing cohort was enriched for high FSIQ scores (mean score > 15 points above population norm of 100; Figure 1a). Population-level FSIQ scores have increased over time, so this may be partially attributed to the use of the same intelligence measures over our study’s multi-decade acquisition period (Flynn, 1987; McDermott et al., 2019), but is important to consider when making comparisons to a clinical population with known cognitive weaknesses. Third, diagnostic and ascertainment biases are known confounds in SCA studies: severe cases are more likely to be clinically recognized and studied (Bardsley et al., 2013; Gunther et al., 2004; van Rijn, 2019), so our sample may not reflect the full phenotypic range of the SCA population. Finally, we included a cross-sectional cohort of participants aged 5–25, but there is strong evidence that the associations between cognition and anatomy are dynamic over developmental timespans (Burgaleta, Johnson, Waber, Colom, & Karama, 2014; Schmitt, Raznahan, et al., 2019; Schnack et al., 2015; Shaw et al., 2006). Future studies in longitudinal cohorts could help determine the most relevant developmental timepoints for SCA-associated cognitive and anatomical disruptions, and provide further insight into the genetic and neurobiological correlates of those changes.

5 |. CONCLUSIONS

Here, we provide the first direct evidence of altered relationships between regional neuroanatomy and cognitive abilities in supernumerary SCA, finding five cortical regions where normative FSIQ–CT relationships were disrupted, and five cortical regions where normative FSIQ–SA relationships were disrupted. We propose that these disturbances—perhaps especially for CT—could underpin known cognitive impairments in SCA, and that sex-chromosomally-mediated genetic influences are key drivers of these changes. These results may be of relevance not only for supernumerary SCA cohorts, but also for other neurogenetic conditions.