In this study, we combined multiple complementary image processing methods, volumetric analyses, VBM, and surface-based anatomic modeling, to demonstrate a discernible neuroanatomic phenotype associated with FraX. Several brain regions were identified as morphologically aberrant in this large group of children with FraX, most prominently, the CN, PV, amygdala, and STG. Of these regions, the CN and PV were correlated with the severity of FMRP deficiency or cognitive deficit in children with FraX. In addition, significant associations were demonstrated between CN volume and severity of aberrant behaviors.
Our hypothesis regarding abnormal CN development and its association to FMRP and severity of cognitive and behavioral symptoms was strongly supported. Gross caudate volumes in male subjects with FraX were more than 20% larger than age-matched male control subjects, and increased size of this region was confirmed in the VBM analyses. The CN enlargement in FraX was especially strong early in development. These results replicate and significantly extend previous findings that showed an association among CN volume, FMR1
gene inactivation (which is highly correlated with FMRP levels), and lower FSIQ.8
Interestingly, increased CN volumes have also been detected in non-FraX children and adults with autism.26–28
In one study of autism, increased CN volumes were associated with repetitive behaviors.27
Furthermore, and in line with results from the surface mapping correlations presented here, enlargement and altered metabolism of the CN in autism have been primarily localized to the right side.27,29
When combined with findings from this study, these results suggest that abnormal neurodevelopment of the CN may be a common pathway leading to a phenotype that includes autistic and stereotypic behaviors.
Our hypothesis regarding the PV in FraX was largely supported. In line with previous studies,7,30
the region of interest analysis indicated that individuals with FraX had significantly reduced PV sizes (approximately 20% in male relative to control subjects). Furthermore, FMRP level positively correlated and FSIQ scores negatively correlated with PV size, suggesting that more abnormal PV development is linked to more severe cognitive deficits in FraX. Though associations between PV size and aberrant behaviors were not as prominent as CN-behavior correlations, decreasing PV size was correlated with increasing scores on the ABC subscale that includes the most items related to stereotypic behaviors (Body and Object Use). This finding is consistent with a previous study of school-age girls with FraX31
where PV size was negatively correlated with stereotypic behaviors, and to a more recent study of non-FraX children with autism32
where the size of lobules VI-VII within the PV was found to be negatively correlated with exploratory and stereotypic behavior.
Our hypotheses regarding the amygdala and STG were partially supported. As hypothesized, both regions were smaller in FraX compared with control subjects, and both were identified by the QROC to distinguish between FraX and control subjects. Portions of the left (GM and white matter) and right STG (white matter only) were also identified as smaller in FraX with VBM. However, no associations were detected between the size of these regions and FMRP levels or FSIQ.
The unanticipated correlation of STG volume with behavioral symptoms is difficult to explain because it occurred in the direction opposite to that expected. However, three of the primary variables of interest, including the STG, independently contributed to cognitive outcome in subjects with FraX in the expected direction (eg, larger STG indicates higher IQ); taken together, the volumes of these regions accounted for more than 30% of the variance in FSIQ. These findings suggest that abnormal neurodevelopment of the CN, PV, and STG contributes to the pathogenesis of cognitive deficits in FraX. These results are also in line with accumulating evidence for the role of these brain structures in higher order cognitive functions including language, visuospatial domains, attention, and executive functions.33,34
Previous studies have suggested that brain morphology in FraX is altered, with particular emphasis on abnormal prefrontal-striatal neuroanatomy.8,13,14
However, the findings of this study provide a more specific picture of abnormal neuroanatomic circuits in FraX. For example, both VBM and surface-based anatomic modeling localized the enlargement of the CN in FraX to lateral and medial parts of the CN head region. The head of the CN is a component of two major corticostriatothalamic circuits.33,35
One circuit connects the dorsolateral prefrontal cortex with the dorsolateral head of CN and is involved in executive functioning. The second circuit connects the lateral orbitofrontal cortex with the ventromedial head of caudate.33,35
Dysfunction of this second circuit is associated with perseverative behavior.35
The clinical phenotype of FraX is highly consistent with impairments of these two corticostriatal circuits as executive functioning deficits and perseverative behaviors are common features of this neurogenetic condition.5,36
This study also provides new evidence for abnormal orbitofrontal-amygdala circuitry in FraX. Although FMR1
knock-out mice demonstrate evidence of amygdala dysfunction including audiogenic seizures and an abnormal conditioned fear response,37,38
to our knowledge, this is the first study to report amygdala and orbitofrontal anatomic abnormalities in a group with the FMR1
full mutation. Orbitofrontal-amygdala circuitry participates in regulation of social behavior and affect.39
Though dysfunction of this circuit may be relevant to clinical features of fragile X such as social anxiety and withdrawal and reduced eye contact,4
we note that, in this study, we did not find an association between amygdala volume and behavioral measures.
The VBM and region of interest approach converged to demonstrate reduced STG volume in FraX, previously shown in a much smaller sample.40
These STG structural abnormalities in FraX are in line with functional magnetic resonance imaging findings showing that subjects with FraX have decreased activation of the superior temporal region during processing of face and gaze stimuli.41
The increase in fusiform gyrus volume detected by the VBM approach might also be related to decreased specificity in activation of this region found in FraX subjects in response to direct gaze.41
Specifically, fusiform enlargement in FraX could be related to lack of maturation and refinement of this face-processing region in affected individuals secondary to gaze aversion and aberrant functional specialization. We are currently collecting specific eye movement data in all new subjects undergoing imaging to further explore this issue.
The lack of complete convergence between VBM and volumetric results in this study could be influenced by tissue segmentation issues. For example, VBM size differences will be more readily identified in structures that have greater tissue homogeneity, such as the CN, than in other regions, such as the amygdala, that contain a greater admixture of GM and white matter.
In this study, we utilized a statistical method (QROC) that addresses limitations of traditional parametric statistical analyses commonly used in analyzing brain imaging data.25,42
In providing a hierarchical profile of neuroanatomy associated with FraX, the QROC identified a combination of large CN, small PV, small amygdala, and small STG that at specific cut-point values distinguished between the FraX and control groups with 92% sensitivity and 61% specificity. These results suggest that QROC may be a valuable tool to identify combinations of brain regions that can distinguish between patients and control subjects.
A potential limitation of our study is aggregating data from different scanners over time, as this raises the question of compatibility of imaging data. Though there were differences in the male/female ratio and mean ages of subjects from the two sites, these differences did not appear to influence the observed between-group findings. The likelihood of a type 1 error resulting from more than one imaging site is also reduced given comparable image characteristics and morphometric results when analyses were broken out by location (see supplementary materials). We also controlled for key image acquisition parameters that are important for maintaining image compatibility across sites.21
Both sites had similar scanner type, slice orientation (coronal), thin slice thickness, field of view, and acquisition matrix size. As shown in a previous study, these parameters are key in preserving image compatibility between sites.21
For the volumetric measures, the approach used for delineating and subdividing the anterior cerebral cortex was limited by the fact that this method did not rely on sulcal/gyral landmarks or on functionally relevant landmarks.
Similarly, other approaches for performing the surface-based modeling procedure could have been used. As a requirement of the method used here, caudate region of interests were first realigned into a standard anatomic space that retained the roughly tubular shape of this structure in the anterior-posterior direction. As an alternative to this method, one could compute the set of points that are the local maxima of three-dimensional distance to the boundary and connect them. A smooth curve could then be computed iteratively with maximal three-dimensional distance to the boundary surface, by numeric optimization or solving boundary-based partial differential equations.43,44
The resulting three-dimensional distance map is intrinsic and does not depend on the orientation of the structure or how it is aligned or sliced. It is not necessarily more accurate to compute the medial curve in this way, but it is rotation invariant and avoids the need to first align the structure into a standard orientation, which may add some small but measurable error associated with rotational errors in registration.
Another limitation is the large cognitive discrepancy between the FraX group and healthy control subjects. Future studies should include cognitively matched control subjects with other forms of developmental disabilities and children with autism (without FraX) to more definitively identify the neuroanatomic phenotype associated with FraX.
Additional brain regions of potential interest to understanding the pathophysiology of brain dysfunction in FraX were also identified. Individuals with FraX in this study demonstrated an aberrant profile of anterior cerebral brain morphology with reduced size of ventral regions and relatively increased size of the dorsal region. The ventrodorsal (smaller to bigger) “gradient” of size in the FraX anterior cerebral cortex is of opposite polarity to that observed in Williams’ syndrome, a neurogenetic condition associated with a contiguous deletion on chromosome 7q.45
This finding is remarkable because individuals with Williams’ syndrome also demonstrate a social behavioral phenotype that is of opposite polarity to that observed in FraX: appetitive as opposed to avoidant. When combined with results showing that frontal and temporal lobes are smaller and parietal and occipital lobes are larger in FraX relative to control subjects, the atypical anterior_cerebral morphology observed here suggests that FMRP may bind to and regulate the translation of messenger RNA important for cortical gradients and shape.
In summary, using comprehensive and complementary analytic approaches, we were able to identify intriguing links between the presence of the FMR1 full mutation and abnormal development of the brain in FraX, in particular the CN and PV. We also demonstrate the potential role of specific brain regions in the pathogenesis of cognitive deficits and aberrant behavior symptoms in FraX. Being a causatively well-defined syndrome, FraX is a compelling model from which to learn about the intricate and complex linkages among measurable factors relating to gene, brain, and cognition.