Disrupted developmental organization of the structural connectome in fetuses with corpus callosum...

NeuroImage 111 (2015) 277–288

Contents lists available at ScienceDirect

NeuroImage

j ourna l homepage: www.e lsev ie r .com/ locate /yn img

Disrupted developmental organization of the structural connectome infetuses with corpus callosum agenesis

András Jakab a,⁎, Gregor Kasprian b, Ernst Schwartz a, Gerlinde Maria Gruber c, Christian Mitter b,Daniela Prayer b, Veronika Schöpf b,1, Georg Langs a,d,1

a Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Computational Imaging Research Lab (CIR) Vienna, Austriab Department for Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Vienna Austriac Center for Anatomy and Cell Biology, Department of Systematic Anatomy, Medical University of Vienna, Vienna, Austriad Computer Science and Artificial Intelligence Lab, Massachusetts Institute of Technology, Cambridge, MA, USA

Abbreviations: CCA, corpus callosum agenesis; DTI, diftionalanisotropy;FDR,falsediscoveryrate;FWER,familywimodel; GW, gestational week; ROI, region of interest; SSE,⁎ Corresponding author at: Computational Imaging Res

Biomedical Imaging and Image-guided Therapy, MWähringer Gürtel 18-20, A1090 Vienna, Austria.

E-mail address: andras.jakab@meduniwien.ac.at (A. Ja1 Contributed equally.

http://dx.doi.org/10.1016/j.neuroimage.2015.02.0381053-8119/© 2015 Elsevier Inc. All rights reserved.

a b s t r a c t

Article history:Accepted 17 February 2015Available online 26 February 2015

Keywords:Corpus callosum agenesisCorpus callosum hypogenesisFetal diffusion MRIDiffusion tensor imagingFetal brain connectivityPrenatal developmentConnectome

Agenesis of the corpus callosum is a model disease for disrupted connectivity of the human brain, in which thepathological formation of interhemispheric fibers results in subtle to severe cognitive deficits. Postnatal studiessuggest that the characteristic abnormal pathways in this pathology are compensatory structures that emergevia neural plasticity. We challenge this hypothesis and assume a globally different network organization of thestructural interconnections already in the fetal acallosal brain.Twenty fetuses with isolated corpus callosum agenesis with or without associatedmalformations were enrolledand fiber connectivity among 90 brain regions was assessed using in utero diffusion tensor imaging and stream-line tractography. Macroscopic scale connectomes were compared to 20 gestational age-matched normally de-veloping fetuses with multiple granularity of network analysis.Gradually increasing connectivity strength and tract diffusion anisotropy during gestation were dominant inantero-posteriorly running paramedian and antero-laterally running aberrant pathways, and in short-range con-nections in the temporoparietal regions. In fetuses with associated abnormalities, more diffuse reduction ofcortico-cortical and cortico-subcortical connectivity was observed than in cases with isolated callosal agenesis.The global organization of anatomical networks consisted of less segregated nodes in acallosal brains, and hubsof dense connectivity, such as the thalamus and cingulate cortex, showed reduced network centrality.Acallosal fetal brains show a globally altered connectivity network structure compared to normals. Besidesthe previously described Probst and sigmoid bundles, we revealed a prenatally differently organizedmacroconnectome, dominated by increased connectivity. These findings provide evidence that abnormal path-ways are already present during at early stages of fetal brain development in the majority of cerebral whitematter.

© 2015 Elsevier Inc. All rights reserved.

Introduction

The corpus callosum is the largest commissural structure in thehuman brain, formed by more than 190 million cross-hemisphericaxons that are known to exhibit excitatory function (Tomasch, 1954;Bloom and Hynd, 2005). Callosal fibers emerge via multiple embryonic,fetal, and postnatal developmental steps that give rise to and sculpt theconnections between cerebral hemispheres (Paul et al., 2007; Richards

fusion tensor imaging; FA, frac-seerrorrate;GLM,general linearsumof squares for errors.earch Lab (CIR), Department ofedical University of Vienna,

kab).

et al., 2004). The classification of developmental brain pathologies iseither based upon the earliest time-point at which a certain develop-mental process (neuronal proliferation, migration, and organization) isdisrupted (Barkovich et al., 2012), or upon the pattern of geneticexpression (Sarnat, 2008). Many brain malformations are primarilycharacterized by abnormal axonal path-finding, regarded as axon-guidance disorders (Engle, 2010). The prenatal disruption of the normalcommissuration can lead to partial or complete agenesis of the corpuscallosum (=callosal agenesis, CCA) a common brain malformationwith a combined prevalence of 0.02–0.5% (Paul et al., 2007; Jeret et al.,1985) in the population and 2–3% of patients with mental retardation(Jeret et al., 1985). CCA, and a wide range of associated neuro-developmental abnormalities, can be assessed by ultrasound examina-tion from mid-gestation (Santo et al., 2012; Vergani et al., 1994;Comstock et al., 1985), although fetal diagnostics becomes challengingfor the fine-grained description of midline structures, as seen in cases

278 A. Jakab et al. / NeuroImage 111 (2015) 277–288

with hypoplastic corpus callosum. To overcome this limitation,recent fast in utero magnetic resonance imaging techniques providevaluable surrogate information for the prenatal counseling of neuro-developmental disorders (Chung et al., 2000; Gowland and Fulford,2004; Brugger et al., 2006; Glenn and Barkovich, 2006; Prayer et al.,2006a; Pugash et al., 2008; Mailath-Pokorny et al., 2012; Tang et al.,2009), and a growing spectrum of imaging sequences can now be ap-plied during the fetal period.

Congenital syndromes thatmanifest in aberrant commissuration aregood models for globally and profoundly impaired neural connectivity.One of the most striking manifestations of the disrupted connectivityis that the interhemispheric information transfer is blocked (Fischeret al., 1992; Quigley et al., 2003; Brown et al., 1999). Therefore, it washypothesized that these patients may suffer from severe functional im-pairments. However, this hypothesis is challenged by the high variabil-ity in the functional phenotype observed in acallosal subjects, and theobservation that cognitive functioning can be, unexpectedly, close tonormal. CCA adds additional complexities due to the high diversity inthe genetic background and the associated syndromes (Bedeschi et al.,2006; Bonneau et al., 2002; Sherr et al., 2005; Edwards et al., 2014), aswell as the theoretical possibility of information transfer through surro-gate pathways that may develop through long-rage neuroplasticity(Fischer et al., 1992; Tovar-Moll et al., 2014; Barr and Corballis, 2002).

The main motivation for our study was to characterize the commonand convergent macroscopic-scale white matter architecture in fetuseswith CCA, prior to the alterations of fiber structure in later life. This earlyexperimental window is of particular importance, as, during infanthoodand adolescence, white matter structure develops rapidly, and thesephysiological changes presumably affect the phenotype of acallosalbrains (Prayer et al., 2006a; Dubois et al., 2013; Giorgio et al., 2008;Luders et al., 2010). The main mechanisms are axonal refinement andpruning during early postnatal development (Innocenti and Price,2005) and experience-driven reorganization in later life (Fields, 2005;Scholz et al., 2009; Lövdén et al., 2010).

Our study aims to investigate callosal agenesis in three different as-pects. First, we choose to use in utero diffusion magnetic resonance im-aging to portray the distributed impairment in macroscopic-scale fiberbundles of the acallosal human brain, and reveal novel whitematter ab-normalities that recur in acallosal fetuses (Kasprian et al., 2008, 2010,2013). In contrast to studies that suggest long-range plasticity as themain mechanism behind the formation of aberrant pathways, such asthe Probst bundle or the sigmoid bundle, we hypothesize that fetuseswith associated and isolated forms of callosal agenesis already show a

Table 1Subject characteristics and imaging findings in fetuses with callosal agenesis. CCA: corpus callo

Identifier Gestational age (week + day) Study group Corpus callosum sta

Fetus 1 22 + 3 2nd trimester group CCA (complete)Fetus 2 22 + 0 2nd trimester group CCA (complete)Fetus 3 23 + 2 2nd trimester group Isolated CCAFetus 4 22 + 6 2nd trimester group Isolated CCAFetus 5 25 + 2 2nd trimester group Isolated CCAFetus 6 24 + 4 2nd trimester group Isolated CCAFetus 7 22 + 2 2nd trimester group CCA (complete)Fetus 8 22 + 0 2nd trimester group Partial CCAFetus 9 35 + 1 3rd trimester group Isolated CCAFetus 10 29 + 0 3rd trimester group Isolated CCAFetus 11 29 + 5 3rd trimester group CCA (complete)Fetus 12 36 + 2 3rd trimester group Isolated CCAFetus 13 34 + 4 3rd trimester group Isolated CCAFetus 14 31 + 6 3rd trimester group Partial CCA (occipitaFetus 15 26 + 2 3rd trimester group CCA (complete)Fetus 16 31 + 3 3rd trimester group CCA (complete)Fetus 17 33 + 0 3rd trimester group Isolated CCAFetus 18 27 + 6 3rd trimester group Isolated CCAFetus 19 27 + 1 3rd trimester group CCA (complete)Fetus 20 29 + 2 3rd trimester group CCA (complete)

substantially altered connectional structure, and this structure isexpressed during early life. Second, it is not yet clear how the patholog-ical white matter structures in axon-guidance disorders emerge andwhether their morphological properties remain stable during prenataldevelopment. We aim to broaden our understanding of the timing ofcallosal malformation by characterizing both the microstructural prop-erties of aberrant white matter structures and picture the general mac-roscopic building principles of the connectome across mid- and lategestation. Finally, complex network analysis of connectivity is used toexplicate the functional integration and segregation in pathologicallywired fetal brains.

Methods

Fetal MRI was clinically indicated to rule out or confirm fetal or pla-cental abnormalities. Prior ultrasound examinations were performedfor initial diagnosis and to assess the gestational age. MRI comprised(1) multi-planar structural MRI examinations with T2-weighted se-quences of the whole fetus and the fetal brain, (2) T1-weighted andecho planar T2-weighted sequences to rule out intracranial hemorrhageand/or blood breakdown products (Prayer et al., 2006b), and (3) surro-gate examinations, such as DTI and resting-state BOLD fMRI, which arenot directly used in clinical decision-making. The study protocol wasapproved by the local ethics committee, the mothers gave written,informed consent prior to the examination, and the research wasconducted according to the principles expressed in the Declaration ofHelsinki.

In utero structural connectivity: cohorts and image acquisition

During the study period of January 2010 toMarch 2014, in utero dif-fusion tensor MRI scans were acquired in 463 fetuses. Partial and com-plete corpus callosum agenesis was diagnosed in 43 cases. Weexcluded cases for the following criteria: bad quality of diffusion tensorimages and excessive fetal motion during the scan (n = 7); extremeventriculomegaly (N15 mm dilatation of the lateral ventricles) or largeinterhemispheric cyst (n = 6); fetal age under 22 gestational weeks(GW, n=4); complex associated developmental abnormality affectingthe brain in multiple lobes (n = 2); or when no ultrasound report andGWassessmentwere available to the institution at the time of image re-trieval (n=4). The final population consisted of 20 fetuses with corpuscallosum agenesis (CCA): two partial and 18 with complete agenesis ofthe corpus callosum, and 11 fetuses with radiologically isolated CCA.

sum agenesis.

tus Associated neuroimaging findings

Pontocerebellar hypoplasia, parietooccipital abnormal gyrificationPontocerebellar hypoplasia, missing cavum septi pellucidi–

–

–

–

Dandy Walker variantMigration abnormality in frontal lobe–

–

Cortical malformation–

–

l parts present) –

Pontocerebellar hypoplasia, parietooccipital abnormal gyrificationPontocerebellar hypoplasia–

–

Abnormal frontomedial sulcationGyration abnormality, interhemispheric cyst

279A. Jakab et al. / NeuroImage 111 (2015) 277–288

The demographics of CCA fetuses and the associated diseases are sum-marized in Table 1. Throughout themanuscript, we refer to our popula-tion as CCA fetuses. This cohort was then divided into two groups: 2nd(n = 8) and 3rd (n = 12) trimester groups in order to lessen the con-founding factor of large anatomical, brain size, and diffusion value dif-ferences. The border of the 2nd and 3rd trimester was defined as the26th postconceptual week. Two control groups were sampled fromthe fetuses with confirmed unaffected brain development and withoutmajor somatic, maternal, or placenta pathologies. An equal number ofcontrols were enrolled in the study (n = 8 and n = 12 for the 2ndand 3rd trimester group, respectively), with the gestational agematched to the CCA groups and considering similar image qualitycriteria for inclusion. The gestational age of the CCA group, at the 2ndtrimester was 23.1 ± 1.2 (range: 22.0 – 25.3), and at the 3rd trimesterwas 31.0 ± 3.3 (range: 26.3 – 36.3). For the controls at 2nd trimester,GA was 23.9 ± 1.1 (range: 22.4 – 25.8), and, for controls at the 3rd tri-mester, GAwas 29.6±2.5 (range: 25.8 – 33.5). Differences in age distri-butions were not significant between the control and CCA groups (2ndtrimester, p=0.19; 3rd trimester, p=0.29). The gestational age of theisolated CCA group (n=11)was 29.4±5 (range: 22.8 – 36.3), while forthe fetuses with associated abnormalities (n=9), the GA was 26.4 ± 4(range: 22 – 31.8); the gestational age did not differ significantly (p =0.137). For investigating isolated CCA and associated CCA fetuses, thesame control group was used as for the entire study cohort (n = 20).

Diffusion tensor magnetic resonance imaging (DTI) was performedon a 1.5 T clinical scanner (Philips Medical Systems, Best, TheNetherlands) using a sensitivity encoding (SENSE) cardiac coil withfive elements (three posterior, two anterior) wrapped around themother's abdomen, utilizing single-shot gradient-recalled echo-planarimaging (EPI) and no cardiac gating. The pregnant women were exam-ined in the supine or left decubitus position (feet first), and no contrastagents or sedatives were administered. In order to receive the optimalMR signal, the coil was readjusted depending on the position of thefetal head during the imaging procedure. All investigations were sched-uled between 7 am and 9 am. Axial slices were positioned orthogonal tothe fetal brainstem. For DTI acquisitions, an axial, single-shot, echo pla-nar imaging sequence was used with a TR of 1745 ms, a TE of 90 ms, anacquisition matrix of 112 × 77 resampled to 256 × 256, a voxel size of0.94 × 0.94 mm, and a slice thickness 3.3 mm without a gap or inter-leaved slices. Depending on the size of the fetal brain and gestationalage, 15–24 slices were acquired, covering the whole brain from thebrainstem to the convexities. Images were acquired using SENSE and apseudo-receive bandwidth of 33 Hz/pixel along the phase encode di-mension, corresponding to a water–fat shift of 6.47 pixel in this direc-tion, and perpendicular to the long axis of the fetal brainstem. Sixteennon-collinear diffusion-weighted magnetic pulsed gradients wereused (one B0 image with a b-value 0 s/mm2 and 15 B1 images with700 s/mm2). Further description, examples, and feasibility of the uti-lized fetal DTI protocol can be found in Kasprian et al. (2008). FetalDTI images were anonymized and transferred to image-processingworkstations in DICOM format.

Fetal diffusion tensor data: general methodological considerations

In utero fetal neuroimaging is an emerging technique that faces nu-merous methodological challenges. To achieve optimal results, weadapted adult and postnatal image processing methods described inthe next paragraphs, and further extended their usability for fetal MRI.The work-flow of characterizing the structural brain network reorgani-zation comprised the following major steps: (1) pre-processing of DTIimages, and fitting of diffusion tensors per each image voxel; (2) deter-ministic tractography-based mapping of the structural connectome;(3) structural network characterization by a graphical, theoreticalapproach; and (4) comparison of CCA and normal cohorts usingnetwork-based statistical testing (Zalesky et al., 2010).

Region of interest system for structural connectivity analysis

All image processing tasks were performed in the space of the origi-nal MRI acquisitions to avoid interpolating data, which is problematicdue to the high variability in fetal anatomy across gestation. Regions ofinterest (ROIs) were based on a custom fetal anatomical atlas consistingof 90 identically sized regions. The construction of the fetal atlas is de-scribed in the Supplementary Document, and anatomical nomenclatureand regions are illustrated in Fig. S1. To map custom ROIs into theindividual image space of subjects, we used two fetal brain anatomicaltemplates: one representative of the 23rd gestational week and one forthe 28th week. Linear, coarse, landmark-based alignment between themean diffusion-weighted and template images was performed, whichwas followed by non-linear warping of images by the f3d algorithm,NiftiREG toolbox (Modat et al., 2010). The inverted transformation (tem-plate to subject space) was used tomap the ROIs into the individual fetalimaging space. Matching accuracy for each subject was observed by anexpert with knowledge of neuroanatomy (A. J.). An overview of thepre-processing and analysis steps of the study is provided in Fig. 1A.

Processing of diffusion tensor images

On the mean diffusion-weighted image of the DTI acquisition, thefetal brain was masked from the surrounding tissue. Masks were delin-eated with a semi-automatic geodesic segmentation approach, imple-mented in the Microsoft GEOS 2.2 for Windows (Criminisi et al., 2008).Minor movements of the fetal head during DTI were corrected usingthe MCFLIRT toolbox implemented in FSL (Jenkinson et al., 2012),where 12-dof (degrees of freedom) affine registrations were performedso that each B1 imagewas alignedwith the first B1 volume, assuming nosignificant motion between the B0 and first B1 volume. Due to the factthat the fetal head moves relative to magnetic gradients defined by theB vectors, it is important to re-adjust each element of the B-vectorwith the corresponding transformation estimate (reorientation of thevector using the rotational component of the transformation matrix).

Next, diffusion tensors were estimated in each voxel. We calculatedscalar parametric maps of the diffusion tensor and performed diffusiontractography using the CAMINO software package (Cook et al., 2006). Ro-bust estimation of the diffusion tensors was done by the RESTORE algo-rithm (Chang et al., 2005), and a single-tensor model was used due tothe low number of gradient directions. Fractional diffusion anisotropymaps (FA) were calculated from the tensorial datasets using the knowngeneral equations (Basser and Pierpaoli, 1996) in the CAMINO package.

Structural connectivity analysis

Whole-brain deterministic diffusion tractographywas initiated fromthe masks covering the fetal white matter, which were generated byeroding the brain masks by four voxels. A fourth-order Runge–Kuttafiber-tracking method (rk4) was utilized with probabilistic nearestneighbor interpolation of streamlines, similar to the interpolation algo-rithm described in (Behrens et al., 2003), while the rk4 tracking ap-proach is detailed in Basser et al. (2000) and further technicaldocumentation is found at: http://cmic.cs.ucl.ac.uk/camino/. For eachseed voxel, 10 iterations were performed, in steps of 0.33 mm (1/10 ofthe slice thickness), and a curve threshold of 75° while tracking wasalso constrained to fractional anisotropy values higher than 0.08.Streamlines and fractional anisotropy values along streamlines werestored for further analysis.

Structural brain connectivity wasmodeled usingwhole-brain deter-ministic tractography. Two different models were provided. First, afteraccessing the subject-aligned 90 ROI system, the connectivity strengthbetween regions ROIi and ROIj is given as:

Si j ¼ logCi j

Vi þ V j

Fig. 1. Overview of image processing and structural connectivity analysis steps. A: A fetal atlas is used to create individualized ROIs in the space of fetal DTI acquisitions. Whole-braintractography is performed and two models of structural connectivity are provided: the number of streamline counts between each region, and fractional anisotropy values along tractsconnecting each region. B: Hypothesis testing on fetal structural connectivity datasets is performed by network-based statistics. In our study, we illustrate results by differential statisticalnetworks: network edges are illustrated where statistical significance is reached.

280 A. Jakab et al. / NeuroImage 111 (2015) 277–288

281A. Jakab et al. / NeuroImage 111 (2015) 277–288

where Cij is the streamline count between regions i and j, and Vi and Vjare the volume of ROIi and ROIj. This correction step was necessary, asinitially all ROIs had equal volume, but after non-linear transformationof atlas space to individual images, they could have undergone aniso-tropic scaling.

Second, average fractional anisotropy alongfiber tracts between ROIiand ROIj FAi,j or FAtract was also used as a measure of connectivitystrength. As a result, for each subject, two 90-by-90 connectivity matri-ces were constructed (Fig. 1B).

Graph theoretical metrics

We calculated the following graph-theoretical metrics to character-ize the regional network topology in CCA and controls (Rubinov andSporns, 2010).

Strength of region i:

ki ¼X

j∈Nwi j

where wij is the connection weight between i and j.The betweenness centrality of region i (Freeman, 1979) is:

bi ¼1

n−1ð Þ n−2ð ÞXh; j∈N

h≠ j; h≠i; j≠i

ρh j ið Þρh j

where ρhj is the number of the shortest paths between h and j,and ρhj(i) is the number of the shortest paths between h and j thatpass through i. N are all network nodes. h and j are the neighboringnodes of i.

The weighted clustering coefficient of the network (Watts andStrogatz, 1998; Onnela et al., 2005) is:

Cw ¼ 1n

Xi∈N

2twiki ki−1ð Þ

where tiw is the weighted geometric mean of triangles around node i:

twi ¼ 12

Xj;h∈N

wi jwihwjh

� �1=3

We calculated the s-core network of the structural brainconnectome using the principles described in Hagmann et al. (2008)and its extension for weighted networks, as described by Alvarez-Hamelin et al. (2006). Further description is provided in the Supple-mentary Document. Within-group reproducibility and edge-space sim-ilarity of the fetal structural connectomes were tested and are reportedin the Supplementary Document, with results illustrated in Fig. S4.

Variability of the connectome

Inter-subject variability of the connectome was demonstrated bycalculating the edge-space similarity of each the connectivity networksacross the population. Assuming that FAij and Sij across subjects corre-spond to the same network edge between the same anatomical regionsi and j, we calculated the edge-space similarity EG,H of connectivitygraphs G and H using the following approach:

EG;H ¼X

d¼0…1DdG;HdX

d¼0…1DdG;H

DdG;H ¼ 2jGd∩Hdj

Gdj j þ Hdj j

where DG,Hd is the Dice coefficient of binary overlap of graphs G and H at

threshold value d, and Gd and Hd are unweighted binary graphs created

by thresholding the graphs G and H at the graph density value of d.Hence EG,H is the index of network similarity that takes values between0 (no similarity) and 1.

Comparing structural connections between study groups

For statistical comparisons of the structural connectomes betweengroups, we utilized the Network-Based Statistics (NBS) method(Zalesky et al., 2010), which was implemented in the NBS Toolbox forMatlab R2014. General linear models (GLM) were formulated inwhich we quantified the inter-group differences in connectivity mea-sures, controlling for the effect of gestational age in the models. Foreach comparison, a permuted GLM analysis with t-test contrasts wasrun for 10,000 iterations. Familywise error rate (FWER)-correctedp values smaller than 0.05 were accepted as significant. For comparingregional descriptors of network connectivity (e.g., regional nodestrength and centrality), we used two-sample t-tests adjusted for falsediscovery rate (FDR). The brain networks were visualized with theBrainNet Viewer for Matlab (Xia et al., 2013).

Results

Clinical diagnostic images and visualizations of aberrant tracts inacallosal cases are provided in Fig. 2. For each CCA subject, we observedthe diffusion anisotropy images and the results of the whole-brain fibertracking, and localized and segmented the bundle of Probst or the sig-moid bundle (Kasprian et al., 2008; Utsunomiya et al., 2006; Wahlet al., 2009). We were able to reveal pathological fiber tracts in 11 ofthe 12 (91.67%) 3rd trimester cases (seven bilaterally and three unilat-erally visible). In a fetus with partial corpus callosum agenesis, we visu-alized the sigmoid bundle exemplarily (Fig. 2B). In the younger group,we had a lower success rate for visualizing the Probst bundle, whichwas visible in five of the eight cases (62.5%), and seen bilaterally infour cases. Individual whole-brain tractography results were markedlydifferent in CCA than in normal subjects (comparison: Fig. 2C andVideo S1).

Impairment of brain connectivity in corpus callosum agenesis during thesecond trimester

Differences between the structural brain networks of CCA and nor-mal fetuses showed a mixed picture of increased and reduced connec-tivity both in the tract-wise fractional anisotropy and in the streamlinecount connectivity models. Results of the structural connectome analy-sis are depicted in Fig. 3. The early alteration of white matter architec-ture was hallmarked by reduced strengths in 24 connections andincreased strength in 45 connections, using the FAtract model and a sta-tistical threshold of t = 2.5 (a description of choosing the optimal teststatistic threshold is described in the manual of the NBS toolbox,under: https://sites.google.com/site/bctnet/comparison/nbs). Similarresults were found using the streamline count model, but affectingmore connections: connectivity strength in 49 network edges was in-creased in CCA,while 51were reduced. Due to the large spatial similaritybetween the altered connections using these two models, we restrictedour qualitative description to the overlapping findings. The spatial pre-disposition of connectivity differences is illustrated in Fig. 3, and inmore detail in Fig. S5. The connectivity increase in CCA was prominentin two different anatomical locations. In the midline, excessive antero-posterior long-fiber connectivity was detected, involving the regions ofthe subcallosal cortex, the anterior cingulate and posterior cingulate,the cuneus in the left hemisphere, and the occipital pole in the righthemisphere. Similar antero-posteriorly oriented, increased connectivitywas found between the subcallosal limbic cortex and the lateral, parie-tal, and temporal cortices, symmetrically in the two hemispheres. Thisfinding was more pronounced in the streamline count model. Second,increased connectivity of short-range tracts was found in the bilateral

Fig. 2. Sample images of fetuses with corpus callosum agenesis. A: Fetus with corpus callosum agenesis at the 30th week of gestation. Left image: coronal T2-weightedMRI illustrates thetypical midline appearance of callosal agenesis. Right images: visualization of the Probst bundle using tractography. B: Fetuswith partial corpus callosum agenesis. Using an approach sim-ilar to the examples, we visualized the sigmoid bundle crossing the midline and connecting the right frontal lobe with the left occipital lobe. C: Comparison of normal and pathologicalwhole-brain fiber tract structure in a CCA fetus (left panel) and a normally developing fetus (right panel) at the 30th week of gestation.

282 A. Jakab et al. / NeuroImage 111 (2015) 277–288

parietal cortex and in the frontal cortex involving the superior frontalgyrus, the middle frontal gyrus, and the right orbitofrontal gyrus.

The absence of the corpus callosum was confirmed by decreasedstreamline count and anisotropy values on the group level. Less stream-lines were found for cross-hemispheric connections, predominantly inthe regions adjacent to the midline (e.g., anterior, middle and posteriorcingulate cortex, cuneus, orbitofrontal gyrus), while latero-lateral re-duction of structural network connectivity was found for edges

involving the superior parietal cortex and the frontal pole. Using thestreamline count model, we detected a diffuse decrease of connectivityin all lobes. Next, we report the averaged FAtract and tract streamlinecount values for the connections (sub-network) that were significantlydifferent between CCA and healthy fetuses. For the tracts of increasedconnectivity: FACCA = 0.19 ± 0.072, FAcontrols = 0.0046 ± 0.044(as non-existing fibers have an FA of 0); SCCA = 3.89 ± 0.58, Scontrols =2.43 ± 1.23 (28-fold difference). For the decreased connections:

Fig. 3.Altered structural brain connectivity in fetuseswith corpus callosum agenesis. A: Network edges illustrate significantlyweaker (top row) and stronger (bottom row) connectivitiesin acallosal brains compared to normally developing subjects. Due to the large similarity between the results using FAtract and streamline count models, herewe depicted the increased ordecreased connectivity patterns that reflect differences between the mean FA values along the fiber tracts, while full visualizations are provided in Figs. S5, S6 and S7. Differentialconnectomes for the two study groups are depicted for the 2nd trimester group (left panels) and 3rd trimester group (middle panels). The statistical test results for the full study popu-lation are visualized in the right panels. Within-lobe connections are depicted according to a color scheme (legend), and thickness of the edges refers to the magnitude of difference be-tween connectivity strength in CCA and healthy fetuses. Grey edges represent inter-modal or inter-lobar connections. Network structure was overlaid on representative anatomicaltemplates of the 23rd and 28th gestational weeks. Anatomical nomenclature is given in Fig. S1. B: Development of aberrant connections across gestation. Connectivity strength wasdepicted using both models in our study, FAtract and logS. As a positive control, we showed that the development of callosal streamline counts in normal subjects was best fitted with asecond-order polynomial function (logS data points: R2 = 0.422, SSE = 4.31); however, this association with the FA over tracts was not significant (FAtract data points: R2 = 0.0975,SSE= 0.0358). Negative controlswere the connections that decreased in CCA (callosal connections), which remained consistently weak across gestation. Aberrantly increased connectiv-ities appear to develop over gestation,with a rapid strengthening that exceeds the rate of thenormal corpus callosumdevelopment, but does not have a definite plateau as does the former.Fitting of a 4th order polynomial function was found to be the most feasible (logS data points: R2 = 0.849, SSE = 2.85; FAtract data points: R2 = 0.266, SSE = 0.0566).

283A. Jakab et al. / NeuroImage 111 (2015) 277–288

FACCA = 0.046 ± 0.034, FAcontrols = 0.21 ± 0.043; SCCA = 3.61 ± 0.69,Scontrols = 4.95± 0.28 (21-fold difference). Results are visualized in de-tail in Fig. S2.

Impairment of brain connectivity in corpus callosum agenesis during thethird trimester

In the third trimester, more connections showed increased strengthin CCA fetuses compared to the second trimester cohort (Fig. 3, middle

images). Corpus callosum agenesis was characterized by reduced fiberstreamline counts in 25 and increased connectivity in 182 regionpairs. FAtract values were decreased in 24 and increased in 192 connec-tions. For optimal visualization in Fig. 3, the statistical threshold wasset to 3.5 in cases of increased network connectivity comparisons.

Increased streamline counts and FAtract values were found foranterio-posteriorly oriented long-range fibers, specifically, comprisingthe midline regions of the cingulate cortex, the precuneus, the cuneusand the fronto-temporal and fronto-occipital connections. Excessive

Fig. 4. Altered structural brain connectivity in fetuseswith isolated and associated callosal agenesis (comparison). A: Network edges illustrate significantlyweaker (top row) and stronger(bottom row) connectivities in acallosal brains compared to normally developing subjects, separately for fetuses with radiologically isolated CCA (left panels) and cases where associatedabnormalities were known (right panels). B: Difference in connectivity strength between isolated and associated cases.

284 A. Jakab et al. / NeuroImage 111 (2015) 277–288

connectivities in CCA were highly symmetrical and similar using eitherFAtract or streamline counts in themodel. Connections running from thesubcallosal cortex and the anterior cingulate cortex to the lateral occip-ital cortex, insula, and parietal cortex also showed increased connectiv-ity strength compared to healthy fetuses. This location suggests that thisexcessive connectivity is the Probst bundle or the superior longitudinalfascicle. Similarly to the younger study group, we detected increasedconnectivity of the short-range connections within the parietal lobe.Reduced connectionswere found inmidline cross-hemispheric connec-tions involving the subcallosal cortex, the cingulate cortices, thethalamus, and the cuneus.

The average connection strengths for the significantly altered con-nections were the following. For the tracts of increased connectivity:FACCA = 0.15 ± 0.065, FAcontrols = 0.012 ± 0.0038; SCCA = 2.19 ±0.91, Scontrols = 0.14 ± 0.066 (112-fold difference). For the decreasedconnections: FACCA = 0.11 ± 0.057, FAcontrols = 0.23 ± 0.066; SCCA =2.98 ± 1.16, Scontrols = 5.32 ± 0.38 (218-fold difference). The alteredconnections are depicted in Fig. 3. Detailed visualization of the connec-tivity differences is given in Fig. S3.

System-level summary metrics of the connectome: strength, integrationand segregation

Anatomical segregation of nodes of the structural connectome wasless pronounced in fetuses with corpus callosum agenesis: while theconnectome was characterized by locally increased streamline counts,regionally increased strength, the mean clustering coefficient was glob-ally reduced. The system-level differences between CCA and normalconnectome were less pronounced in the 2nd trimester group. Al-though NBS revealed increased streamline count in a variety of connec-tions in CCA, during the 2nd trimester, the summary metric of nodal

strength was not significant between the two groups after correctingfor multiple comparisons. Without adjusting the p-values for multiplecomparisons, the most significant increase of nodal strengths in CCAwas found for the left superior temporal gyrus (puncorrected = 0.0147),right lateral occipital cortex (puncorrected = 0.0188) and the right occip-ital pole (puncorrected= 0.0108). Using the FAtract connectionmodel, CCAwas associated with reduced regional betweenness centrality in the leftposterior insula (bCCA = 51 ± 66.87; bcontrols = 281.5 ± 143.92), theright lateral thalamus (bCCA = 40.75 ± 38.97; bcontrols = 559.25 ±367.52), and the posterior thalamus (bCCA = 56.25 ± 70.33; bcontrols =477± 103.24, pcorrected b 0.05). The structural network of 2nd trimesterCCA fetuses showed decreased anatomical segregation of connectivities,whichwas indicated by the significantly smaller mean clustering coeffi-cient (Rubinov and Sporns, 2010) in the streamline count model(CCA = 1.43 ± 0.42; Ccontrols = 2.02 ± 0.21, p = 0.0033).

In the third trimester, we found significantly increased strengths incallosal agenesis in 11 brain regions when using the streamline countconnectivity model and 30 regions when using the FAtract model(pcorrected b 0.05). Differences were found for bilateral occipital poles,as well as the bilateral postcentral gyrus. There was an excess of in-creased right hemispheric connectivity, affecting the right precuneus,the right callosal cortex, the right temporal lobe (superior, middle tem-poral gyri, fusiform area), and the right insula. The average increase ofnodal strength was 125.9 ± 47% (range: 56.6 – 254.8%). No nodesshowed significantly reduced connectivity strength. Betweenness cen-trality was reduced in the left supramarginal gyrus region and in theposterior thalamus (streamline count model and in seven regionsusing the FAtract model: the leftmiddle cingulate, the right posterior cin-gulate, the left and right thalamus regions, and the right pallidum. Thereduction in the betweenness centrality measure was 72.6 ± 12.9%(range: 53.2 – 86.8). The structural connectome of 3rd trimester brains

285A. Jakab et al. / NeuroImage 111 (2015) 277–288

with corpus callosumagenesiswas characterized by a smaller clusteringcoefficient using the streamline count model (CCA = 1.61 ± 0.42;Ccontrols = 1.99 ± 0.26, p = 0.0012).

The summary metrics in CCA and normals are plotted in Fig. 5C. Thedensest sub-networks of the connectivity graphs were explicated withs-core network decomposition; this experiment is reported in the Sup-plementary Document and illustrated in Fig. S4.

The brain connectome in fetuses with isolated and associated CCA

In fetuses where the radiological work-up did not reveal develop-mental disorders besides CCA, the reduced connectivitywasmostly lim-ited to interhemispheric fibers, with an exception for the isolatedconnectivity edges in the temporal and frontal lobes. In comparison(Fig. 4), in fetuseswith associatedmalformations,more frontal and tem-poral cortico-cortical connections had reduced streamline counts ortract diffusion anisotropy. Isolated CCA cases were observed to have amore symmetric structure of excessive fronto-occipital and temporo-occipital connectivity, while, in the associated subjects, the aberrantlyincreased connectivity was limited to midline long connections. Resultsfor this analysis are depicted in Figs. S5 and S6.

Variability of the connectome in CCA and normal fetuses

CCA and healthy cases formed clusters within the similarity matrix,which refers to high within-group similarity and marked betweengroup differences. The connectome of CCA fetuses was more variablethan that of controls, and the variability decreased by gestational agein both groups. Results of the network similarity analysis are depictedin Fig. 5A. The coefficient of variation confirmed these results: thepopulation-level variability of connections was higher in CCA subjectsthan in normally developing fetuses (Fig. 5B). The highest edge-spacesimilarities were found for normal 3rd trimester subjects (FAtract

model, Eg,h = 0.34 ± 0.05, streamline count model, Eg,h = 0.54 ±

Fig. 5. Variability of connectomes and network values in CCA and normal fetuses. A: Similarity orepresents the similarity between two subjects. The rows of the matrix have been ordered to sgiven as the Dice coefficients between any two binary-thresholded connectivity matrices, whichument). Left image: similarity using streamline counts model; Right image: similarity of connetions. Each data point represents a connection, plotted according to the inter-subject standard dstrength and betweenness centrality, separately listed for CCA and normal subjects.

0.08). For normal subjects in the earlier GW group (2nd trimestergroup), the graph similarity was lower (FAtract model, Eg,h = 0.3 ±0.04, streamline count model, Eg,h = 0.48 ± 0.05). Similarity of theconnectome of CCA subjects was small in the 2nd trimester (FAtract

model, Eg,h = 0.23 ± 0.07, streamline count model, Eg,h = 0.34 ±0.09) and higher in the 3rd trimester (FAtract model, Eg,h = 0.31 ±0.05, streamline count model, Eg,h = 0.49± 0.07), but this is still signif-icantly lower than the similarity metrics of normal subjects, thereforerepresenting higher variability.

Age-related changes of the aberrant connectome of acallosal brains

We noted the following differences in the connectome associatedwith CCA between early and late gestational ages. A prominent increasein aberrant connectivity is seen at late gestation between the occipitalpole and insula-frontal regions in CCA (Fig. 3A). Furthermore, symmet-ric, long-range latero-lateral connections are more prominent in olderfetuses, and these do not cross the midline. The difference betweenCCA and normal fetuses is more pronounced at later gestational ages(increase of fiber streamline counts is 21-fold in CCA at the early gesta-tion, while it is 112-fold in the older cohort).

Fig. 3B depicts the age-related changes in the impaired connectivi-ties of acallosal brains. As a positive control, we observed the develop-ment of callosal connections in normal subjects. These connectionsremained very sparsely interconnected in acallosal subjects (negativecontrol). The excessive connections in CCA showed a marked develop-mental trajectory: from GW 22nd until GW 32nd, the rate of connectiv-ity increasewas very steep, exceeding the normalmaturation rate of thecorpus callosum. At the end of the observed GW range, the aberrantconnectivities became as strong as the callosal connections in normalsand the rate of strengthening became lower. Fig. 3B demonstratesthat, while connectivity strength as logS is stable, FAtract values show alarge scatter around the fit curves, and therefore, are less reliable. Thenormal callosal connection development was best fitted with a

f fetal structural connectivity networks in the study population. Each element in thematrixeparate fetuses according to the trimester of gestation and the study groups. Similarity isare integrated along structural cost thresholds 0–1 (described in the Supplementary Doc-ctivity using the FAtract model. B: Population-level coefficient of variation (CV) of connec-eviation and the mean connection strength. C: The ten top brain nodes in terms of highest

286 A. Jakab et al. / NeuroImage 111 (2015) 277–288

second-order polynomial function with a maximum point aroundGW 32. The development of aberrantly increased connections wasfitted with a 4th order polynomial function, which is characterized bysteep increase initially and then a relative slow-down. The effect of ges-tational age on the overall structural connectivity is described in theSupplementary Document and illustrated in Fig. S7.

Discussion

In this study, we investigated structural and functional characteris-tics of late 2nd and 3rd trimester fetuses with callosal agenesis usingin utero DTI. The ongoing improvement in prenatal imaging techniques(Brugger et al., 2006; Prayer et al., 2006a; Kasprian et al., 2008, 2013;Schöpf et al., 2012; Bui et al., 2006; Mitter et al., 2011) has resulted inthe ability to identify and visualize fetal brain anomalies before theage of human viability (24 gestational weeks). Recently, insights gainedbyneuroimaging and genetics have led to the identificationof a growingnumber of disorders, characterized by abnormal axon guidance (Engle,2010). As many of these conditions—such as callosal agenesis—are al-ready phenotypically expressed during fetal life, the challenge for futureprenatal medicine is not only to detect but to evaluate these cases(Kasprian et al., 2013).

Our results showed that structural connectivity of CCA fetuses wasmarkedly different from normally developing controls, characterizedby a mixture of increased (excessive) and decreased connectivity andlocal changes in brain regions that normally serve as connectionalhubs (Hagmann et al., 2008). Decreased connectivity was dominatedby the lack of midline interhemispheric connections, matching the pat-tern that was expected from impaired commissuration. This midlinepattern of aberrant white matter structures remained constant overthe period of gestation, comprising interhemispheric connections, butalso shorter-range connections in the polar parts of the frontal lobe. Infetuses with associated developmental abnormalities, the phenotypewas dominated by more diffusely reduced connectivity and less sym-metric excessive fiber pathways.While at the end of the second trimes-ter, the aberrant connections appear to bemuchweaker than the corpuscallosum of healthy controls, and after the 30th and 32nd GW, theygrow to be equally strongly interconnected at 35th GW (Fig. 3).

Neuroanatomical underpinnings of corpus callosum agenesis

Heterotopic fibers persisting in later life are widely accepted to befundamental to axon-guidance disorders (Engle, 2010). Rerouting ofcallosal pathwaysmay result into heterotopic axonal pathways or thick-ening of preexisting tracts or phylogenetically older commissures. Themost common macroscopically observable heterotopic tracts are thebundles of Probst (Probst, 1901), which recently have been proven byprenatal DTI (Kasprian et al., 2013). From adult neuroimaging studiesand case reports, it appears that the Probst bundle has an internalside-by-side topology (Utsunomiya et al., 2006; Tovar-Moll et al.,2007) and probably conveys functional signals (Lefkowitz et al., 1991).Thickening or enlargement of the anterior and hippocampal commis-sure has been suspected to be the result of increased axonal connectiv-ity in a series of studies. Two additional, distinct bundles were reportedin Tovar-Moll et al. (2014), connecting contralateral, presumably func-tioning parietal regions through surrogate pathways in the anteriorand posterior commissures. Moreover, Kasprian et al. reported the in-crease of diffusion anisotropy of the corticospinal tracts in DTI imagesof fetuses with CCA (Kasprian et al., 2013), which could be an indirectsign of increased axon counts or denser axonal packagingwithin an im-aged voxel (Beaulieu, 2002). Finally, the sigmoid bundle connecting thefrontal and contralateral occipital lobes, amore diverse set of interhemi-spheric abnormal pathways were described in partial agenesis of thecorpus callosum (Wahl et al., 2009).

Our study provides evidence toward further aberrant connectivity inCCA; however, the tracts underpinning the increased connectivity

might be eliminated in later life (Innocenti and Price, 2005), whichwould explain the lack of their presence in postnatal studies and thelack of works describing increased diffusion anisotropy in adults.Kasprian et al. confirmed the aberrant trajectory of the sigmoid andProbst bundle (Kasprian et al., 2013), and, supporting ourfindings of ex-cessive connectivity, increased FA values were detected for somatosen-sory and motor fiber tracts. The corpus callosum is already formed bythe 20th gestational week (Raybaud, 2010); however, axonal growthcontinues until the 31st week (Luttenberg, 1966; Jovanov-Milosevicet al., 2006). We confirmed that callosal fiber anisotropy also follows asecond-order polynomial growth with a maximum point (Fig. 3B) anda periodwhere growth slows down or appears to stop due to the retrac-tion of callosal fibers (Clarke et al., 1989; Harreld et al., 2011). In con-trast, the apparent emergence of aberrant fiber connectivities followsa faster trajectory, and the slow-down period is not as prominent asfor the normal callosal fibers.

The structural connectome in corpus callosum agenesis

Similar to our network-level investigation, Owen et al. used graphanalysis to compare the structural connectivity between acallosal casesand neurotypical adults (Owen et al., 2013). The absence of inter-hemispheric fibers was associated with the pervasive re-organizationof cortical and subcortical connectivity that can only partially be attrib-uted to the well-known aberrant long callosal fibers of the Probstbundle. In their study, acallosal brains showed globally preserved con-nectivity degree distribution, which is indirect evidence that ectopictracts maintain the density and cost of the overall brain network, al-though at the expense of reduced centrality and the weakening of themidline structural core of the brain (Hagmann et al., 2008). In contrast,we confirmed preserved network strength only in the early cohort offetuses between gestational weeks 22 and 25. For the study grouprepresenting fetuses older than 26th GW, increased regional networkstrengths indicate exuberant—or excessive—fiber communities. We re-ported similar aberrantmidline tracts that only partially match the clas-sic trajectory of the Probst bundle (comparison: Fig. 9 in (Owen et al.,2013)). In linewith ourfindings on edge-space similarity of connections,Owen et al. found that the CCA connectomehas lower consistency acrosssubjects (Fig. 5A). Acallosal brains appear to have less anatomical segre-gation of its connections, indicated by decreased clustering coefficient.While the central hubs of connectivity are the thalamus and posteriorcingulate in normal fetuses, in the acallosal brains, the densest connec-tivity and largest centrality are found in the cingulate only and thesubcallosal cortex. The alternative organization of intra-hemisphericconnectivities is also marked by increased centrality and connectionstrength in the occipital lobe in CCA fetuses.

Prenatal origin of the aberrant connectivities in acallosal brains:possible explanations

We suggest two hypotheses for the observed excessive connectivityin acallosal fetuses. It is known that, in case of aberrantmidline pattern-ing during development, callosal axons that originate near the cingulatecortex or paramedian regions may turn anteriorly or posteriorly,forming aberrant long fibers bilaterally. We speculated that while thismechanism gives rise to the Probst bundle adjacent to the impairedmidline regulators of commissuration, interhemispheric fibers furtheraway may not reach the midline and fail to form longitudinal fibers.Tovar-Moll et al. provided a similar hypothesis for their description ofsurrogate pathways through the large commissures (Tovar-Moll et al.,2014); however,we cannot confirm this due to fetal DTI's inability to re-solve the anterior or posterior commissural fibers during tractography.Normally, we saw the moderate increase of fiber connectivity in thehealthy corpus callosum,which ratewas exceeded by the strengtheningof the aberrantly increased connections. This first description ofthis phenomenon could reflect the fact that aberrant connections are

287A. Jakab et al. / NeuroImage 111 (2015) 277–288

accumulated during gestation, probably due to long-range plasticity, assuggested previously (Tovar-Moll et al., 2007, 2014; Staudt, 2010). Bythe same token, it has been suggested before that lateral callosal fiberscan join corticospinal axons and result in a larger capsula interna,which would also cause the relative increase of FA in this region, or inother, association tracts (Sarnat, 2008).

Second, conventional diffusion tensor imaging limits thedepiction ofcallosal fibers further away from the midline, mainly due to the fiber-crossing problem during tractography approaches (Behrens et al.,2007). With the relatively few diffusion-encoding gradient directionsused in our study, we should consider the possibility that the missingcallosal fibers artificially unmask intrahemispheric connectivities.This might be pronounced in regions near the cerebral convexitiesand areas where the projection, association, and callosal fibers areintertwined. Image areas with multiple crossing fiber populations,such as seen in the trigonum area in adults or within the superior longi-tudinal fascicle, may falsely contain voxels of low fractional anisotropyand deterministic fiber tracts are thus truncated. It is possible that thedevelopmental absence of callosal connectivity in such areas artificiallyincreases the fractional anisotropy and alters structural connectivitystrengths. Imaging techniques with finer angular resolution orQ-space sampling would overcome this problem (Wahl et al., 2009), al-though their adaptation to living human fetuses is not yet clinically pos-sible. If the observed exuberance of connections in CCA is a crossing-fiber artifact, the gradual increase of the fiber connectivity seen inFig. 3B leads to interesting implications. In this case, it can be further as-sumed that callosalfibers that donot reach themidline are continuouslyeliminated during gestation, causing the observed gradual unmaskingof dominant normal connectivity in the fetal brain in DTI scans, as thegrowth of FA values within aberrant fibers seems to be faster thanthat of the normally developing corpus callosum (Fig. 3B).

Limitations

Our prenatal neuroimaging study suffers from the following limita-tions. During diffusion MRI, motion against the external field B1 causesspin dephasing and reduces the signal-to-noise ratio, calling for specificattention when interpreting fetal diffusion anisotropy or mean diffusiv-ity values. We compared FAtract values across groups acquired undersimilar conditions, lessening this effect for the reported group differ-ences. We used a manually defined ROI system to map structural con-nectivity, which is not optimal due to the unknown correspondence ofthe macroscopical appearance of the fetal cortex and the underlyingfunctional activity. To the best of our knowledge, no digital corticalatlases exist for fetuses; therefore, wemust rely onmanual delineationsor data-driven parcellations of our data. Achieving accurate correspon-dences across subjects is also limited due to the rapidly changingbrain anatomy during gestation. Differences in the brain morphology,especially in brainswith pronounced colpocephaly can represent a con-founding factor when comparing CCA and normally developingsubjects.

The reproducibility of our study is confounded by small case num-bers. Themain limiting factor to increasing case numbers is the high ex-clusion rate due to artifacts and the strict indications to perform fetalMRI. In our study design, CCA is a concatenated group of abnormally de-veloping fetuses with partial or total agenesis of the corpus callosumandmanywere diagnosedwith other malformations aswell. Clusteringthe study cohort into more homogeneous groups is, however, not yetpossible due to the low case numbers and the fact that the genetic back-ground behind CCA is very heterogeneous, with callosal agenesis beingonly the tip of the iceberg of the underlying abnormality. Therefore, theprecise determination of the relationship between genotype, etiology,and the structural–functional phenotype of this disease would requirelarger study groups in the future.

The acquisition protocol for fetal DTI suffers from low angular reso-lution; with the current settings, it is not possible to sufficiently model

more than one fiber population per voxel. In fetal imaging, longer scantime during DTI will greatly increase the ratio of corrupted imageframes due to fetal headmotion andmaternal breathing, and this limitsthe number of applicable diffusion-weighting directions. The more ac-curate modeling of the diffusion propagator, and the application ofprobabilistic tractography techniques to the fetal brain would result inmore reproducible description of the structural connectome in the fu-ture (Behrens et al., 2003, 2007).

Besides the technical limitation, the exact microstructural correlatesof the “fibers” revealed in our study is unknown. Previous studies re-vealed correspondence between the in utero DTI based visualizationof the Probst bundle and histological findings (Kasprian et al., 2013);however, we do not yet know the substrates of the newly described ab-errant interconnections. Before gestational week 24, neither neuronalmigration, axonal growth, pathfinding, nor myelination is complete.Thus, other transient cellular components, such as radial glia and eventhemigrational “coherence” of neuronsmay contribute to and confoundin utero fetal DTI findings.

Conclusions

Misguided axons in corpus callosum agenesis form aberrant path-ways not only adjacent to the median surface of the hemispheres, butdistributed across the lateral parts of the convexity. In fetuses, we haveshown by connectome-level tractography that callosal agenesis mani-fests in excessive structural connectivity, which constantly intensifiesduring gestation. The maturation of aberrant callosal pathways followsa trajectory similar to that of normal interhemispheric connectivity,but exceeding that in the speed of increasing structural connectivitystrength and fiber diffusion anisotropy. Our findings provide evidencethat the abnormal fibers in CCA aremore governed by genetically deter-mined prenatal events than later-life compensatory processes.

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.neuroimage.2015.02.038.

Funding

This work was supported by grants from the Austrian National BankAnniversary Fund (14812 to G.K. and 15929 to V.S.), the EU FP7 projectsKHRESMOI (FP7-ICT-2009-5/257528) to G.L. A. J. is supported by theEuropeanUnion FP7Marie Curie IEF Research grant FABRIC—“Exploringthe Formation and Adaptation of the Brain Connectome,” grant no.2012-PIEF-GA-33003.

Acknowledgments

The authors thank Dr. Mariella Polterauer and Prof. Dieter Bettel-heim for clinical contributions.

References

Alvarez-hamelin, J.I., Barrat, A., Vespignani, A., 2006. Large Scale Networks Fingerprintingand Visualization Using the k-Core Decomposition. pp. 41–50.

Barkovich, A.J., Guerrini, R., Kuzniecky, R.I., Jackson, G.D., Dobyns, W.B., 2012. A develop-mental and genetic classification for malformations of cortical development: update2012. Brain 135, 1348–1369.

Barr, M.S., Corballis, M.C., 2002. The role of the anterior commissure in callosal agenesis.Neuropsychology 16, 459–471.

Basser, P.J., Pierpaoli, C., 1996. Microstructural and physiological features of tissues eluci-dated by quantitative-diffusion-tensor MRI. J. Magn. Reson. Ser. B 111, 209–219.

Basser, P.J., Pajevic, S., Pierpaoli, C., Duda, J., Aldroubi, A., 2000. In vivo fiber tractographyusing DT-MRI data. Magn. Reson. Med. 44, 625–632.

Beaulieu, C., 2002. The basis of anisotropic water diffusion in the nervous system? A tech-nical review. NMR Biomed. 15, 435–455.

Bedeschi, M.F., Bonaglia, M.C., Grasso, R., Pellegri, A., Garghentino, R.R., et al., 2006.Agenesis of the corpus callosum: clinical and genetic study in 63 young patients.Pediatr. Neurol. 34, 186–193.

Behrens, T.E.J., Woolrich, M.W., Jenkinson, M., Johansen-Berg, H., Nunes, R.G., et al., 2003.Characterization and propagation of uncertainty in diffusion-weighted MR imaging.Magn. Reson. Med. 50, 1077–1088.

288 A. Jakab et al. / NeuroImage 111 (2015) 277–288

Behrens, T.E.J., Berg, H.J., Jbabdi, S., Rushworth, M.F.S., Woolrich, M.W., 2007. Probabilisticdiffusion tractography with multiple fibre orientations: what can we gain?Neuroimage 34, 144–155.

Bloom, J., Hynd, G., 2005. The role of the corpus callosum in interhemispheric transfer ofinformation: excitation or inhibition? Neuropsychol. Rev. 15, 59–71.

Bonneau, D., Toutain, A., Laquerriere, A., Marret, S., Saugier-Veber, P., et al., 2002. X-linkedlissencephaly with absent corpus callosum and ambiguous genitalia (XLAG): clinical,magnetic resonance imaging, and neuropathological findings. Ann. Neurol. 51,340–349.

Brown, W.S., Jeeves, M.A., Dietrich, R., Burnison, D.S., 1999. Bilateral field advantage andevoked potential interhemispheric transmission in commissurotomy and callosalagenesis. Neuropsychologia 37, 1165–1180.

Brugger, P.C., Stuhr, F., Lindner, C., Prayer, D., 2006. Methods of fetal MR: beyondT2-weighted imaging. Eur. J. Radiol. 57, 172–181.

Bui, T., Daire, J.L., Chalard, F., Zaccaria, I., Alberti, C., et al., 2006. Microstructural develop-ment of human brain assessed in utero by diffusion tensor imaging. Pediatr. Radiol.36, 1133–1140.

Chang, L.C., Jones, D.K., Pierpaoli, C., 2005. RESTORE: robust estimation of tensors by out-lier rejection. Magn. Reson. Med. 53, 1088–1095.

Chung, H.W., Chen, C.Y., Zimmerman, R.A., Lee, K.W., Lee, C.C., et al., 2000. T2-weightedfast MR imaging with true FISP versus HASTE: comparative efficacy in the evaluationof normal fetal brain maturation. AJ. Am. J. Roentgenol. 175, 1375–1380.

Clarke, S., Kraftsik, R., Van der Loos, H., Innocenti, G.M., 1989. Forms andmeasures of adultand developing human corpus callosum: is there sexual dimorphism? J. Comp.Neurol. 280, 213–230.

Comstock, C.H., Culp, D., Gonzalez, J., Boal, D.B., 1985. Agenesis of the corpus callosum inthe fetus: its evolution and significance. J. Ultrasound Med. 4, 613–616.

Cook PA, Bai Y, Nedjati-Gilani SK, Seunarine K, Hall MG, et al. (2006).Criminisi, A., Sharp, T., Blake, A., 2008. Geos: geodesic image segmentation. Anonymous

Computer Vision—ECCV 2008. Springer, pp. 99–112.Dubois, J., Dehaene-Lambertz, G., Kulikova, S., Poupon, C., Huppi, P.S., et al., 2013. The

early development of brain white matter: a review of imaging studies in fetuses,newborns and infants. Neuroscience 276, 48–71.

Edwards, T.J., Sherr, E.H., Barkovich, A.J., Richards, L.J., 2014. Clinical, genetic and imagingfindings identify new causes for corpus callosum development syndromes. Brain 137,1579–1613.

Engle, E.C., 2010. Human genetic disorders of axon guidance. Cold Spring Harb. Perspect.Biol. 2, a001784.

Fields, R.D., 2005. Myelination: an overlooked mechanism of synaptic plasticity? Neuro-scientist 11, 528–531.

Fischer, M., Ryan, S.B., Dobyns, W.B., 1992. Mechanisms of interhemispheric transfer andpatterns of cognitive function in acallosal patients of normal intelligence. Arch.Neurol. 49, 271–277.

Freeman, L.C., 1979. Centrality in social networks conceptual clarification. Soc. Networks1, 215–239.

Giorgio, A., Watkins, K.E., Douaud, G., James, A.C., James, S., et al., 2008. Changes in whitematter microstructure during adolescence. Neuroimage 39, 52–61.

Glenn, O.A., Barkovich, A.J., 2006. Magnetic resonance imaging of the fetal brain andspine: an increasingly important tool in prenatal diagnosis, part 1. AJNR Am.J. Neuroradiol. 27, 1604–1611.

Gowland, P., Fulford, J., 2004. Initial experiences of performing fetal fMRI. Exp. Neurol. 190(Supplement 1), 22–27.

Hagmann, P., Cammoun, L., Gigandet, X., Meuli, R., Honey, C.J., et al., 2008. Mapping thestructural core of human cerebral cortex. PLoS Biol. 6, e159.

Harreld, J.H., Bhore, R., Chason, D.P., Twickler, D.M., 2011. Corpus callosum length by ges-tational age as evaluated by fetal MR imaging. AJNR Am. J. Neuroradiol. 32, 490–494.

Innocenti, G.M., Price, D.J., 2005. Exuberance in the development of cortical networks. Nat.Rev. Neurosci. 6, 955–965.

Jenkinson, M., Beckmann, C.F., Behrens, T.E., Woolrich, M.W., Smith, S.M., 2012. Fsl.Neuroimage 62, 782–790.

Jeret, J.S., Serur, D., Wisniewski, K., Fisch, C., 1985. Frequency of agenesis of the corpuscallosum in the developmentally disabled population as determined by computer-ized tomography. Pediatr. Neurosci. 12, 101–103.

Jovanov-Milosevic, N., Benjak, V., Kostovic, I., 2006. Transient cellular structures in devel-oping corpus callosum of the human brain. Coll. Antropol. 30, 375–381.

Kasprian, G., Brugger, P.C., Weber, M., Krssák, M., Krampl, E., et al., 2008. In uterotractography of fetal white matter development. Neuroimage 43, 213–224.

Kasprian, G., Del Rio, M., Prayer, D., 2010. Fetal diffusion imaging: pearls and solutions.Top. Magn. Reson. Imaging 21, 387–394.

Kasprian, G., Brugger, P.C., Schopf, V., Mitter, C., Weber, M., et al., 2013. Assessing prenatalwhite matter connectivity in commissural agenesis. Brain 136, 168–179.

Lefkowitz, M., Durand, D., Smith, G., Silver, J., 1991. Electrical properties of axons withinprobst bundles of acallosal mice and callosi that have reformed upon glial-coatedpolymer implants. Exp. Neurol. 113, 306–313.

Lövdén, M., Bodammer, N.C., Kühn, S., Kaufmann, J., Schütze, H., et al., 2010. Experience-dependent plasticity of white-matter microstructure extends into old age.Neuropsychologia 48, 3878–3883.

Luders, E., Thompson, P.M., Toga, A.W., 2010. The development of the corpus callosum inthe healthy human brain. J. Neurosci. 30, 10985–10990.

Luttenberg, J., 1966. Contribution to the fetal ontogenesis of the corpus callosum in man.3. Myelinization in the corpus callosum. Folia Morphol. (Praha) 14, 192–199.

Mailath-Pokorny, M., Kasprian, G., Mitter, C., Schöpf, V., Nemec, U., et al., 2012. Magneticresonance methods in fetal neurology. 17, 278–284.

Mitter, C., Kasprian, G., Brugger, P.C., Prayer, D., 2011. Three-dimensional visualization offetal white-matter pathways in utero. Ultrasound Obstet. Gynecol. 37, 252–253.

Modat, M., Ridgway, G.R., Taylor, Z.A., Lehmann, M., Barnes, J., et al., 2010. Fast free-formdeformation using graphics processing units. Comput. Methods Programs Biomed.98, 278–284.

Onnela, J., Saramaki, J., Kertesz, J., Kaski, K., 2005. Intensity and coherence of motifs inweighted complex networks. Phys. Rev. E 71, 065103.

Owen, J.P., Li, Y.O., Ziv, E., Strominger, Z., Gold, J., et al., 2013. The structural connectome ofthe human brain in agenesis of the corpus callosum. Neuroimage 70, 340–355.

Paul, L.K., Brown, W.S., Adolphs, R., Tyszka, J.M., Richards, L.J., et al., 2007. Agenesis of thecorpus callosum: genetic, developmental and functional aspects of connectivity. Nat.Rev. Neurosci. 8, 287–299.

Prayer, D., Kasprian, G., Krampl, E., Ulm, B., Witzani, L., et al., 2006a. MRI of normal fetalbrain development. Eur. J. Radiol. 57, 199–216.

Prayer, D., Brugger, P.C., Kasprian, G., Witzani, L., Helmer, H., et al., 2006b. MRI of fetalacquired brain lesions. Eur. J. Radiol. 57, 233–249.

Probst, M., 1901. Ueber den Bau des vollständig balkenlosen Grosshirnes sowie überMikrogyrie und Heterotopie der grauen Substanz. Arch. Psychiatr. Nervenkr. 34,709–777.

Pugash, D., Brugger, P.C., Bettelheim, D., Prayer, D., 2008. Prenatal ultrasound and fetalMRI: the comparative value of each modality in prenatal diagnosis. Eur. J. Radiol.68, 214–226.

Quigley, M., Cordes, D., Turski, P., Moritz, C., Haughton, V., et al., 2003. Role of the corpuscallosum in functional connectivity. AJNR Am. J. Neuroradiol. 24, 208–212.

Raybaud, C., 2010. The corpus callosum, the other great forebrain commissures, and theseptum pellucidum: anatomy, development, and malformation. Neuroradiology 52,447–477.

Richards, L.J., Plachez, C., Ren, T., 2004. Mechanisms regulating the development of thecorpus callosum and its agenesis in mouse and human. Clin. Genet. 66, 276–289.

Rubinov, M., Sporns, O., 2010. Complex networkmeasures of brain connectivity: uses andinterpretations. Neuroimage 52, 1059–1069.

Santo, S., D'Antonio, F., Homfray, T., Rich, P., Pilu, G., et al., 2012. Counseling in fetalmedicine: agenesis of the corpus callosum. Ultrasound Obstet. Gynecol. 40, 513–521.

Sarnat, H.B., 2008. Embryology and malformations of the forebrain commissures. Handb.Clin. Neurol. 87, 67–87.

Scholz, J., Klein, M.C., Behrens, T.E.J., Johansen-Berg, H., 2009. Training induces changes inwhite-matter architecture. Nat. Neurosci. 12, 1370–1371.

Schöpf, V., Kasprian, G., Brugger, P.C., Prayer, D., 2012. Watching the fetal brain at ‘rest’.Int. J. Dev. Neurosci. 30, 11–17.

Sherr, E.H., Owen, R., Albertson, D.G., Pinkel, D., Cotter, P.D., et al., 2005. Genomic micro-array analysis identifies candidate loci in patients with corpus callosum anomalies.Neurology 65, 1496–1498.

Staudt, M., 2010. Reorganization after pre- and perinatal brain lesions. J. Anat. 217,469–474.

Tang, P.H., Bartha, A.I., Norton, M.E., Barkovich, A.J., Sherr, E.H., et al., 2009. Agenesis of thecorpus callosum: an MR imaging analysis of associated abnormalities in the fetus.AJNR Am. J. Neuroradiol. 30, 257–263.

Tomasch, J., 1954. Size, distribution, and number of fibres in the human corpus callosum.Anat. Rec. 119, 119–135.

Tovar-Moll, F., Moll, J., de Oliveira-Souza, R., Bramati, I., Andreiuolo, P.A., et al., 2007.Neuroplasticity in human callosal dysgenesis: a diffusion tensor imaging study.Cereb. Cortex 17, 531–541.

Tovar-Moll, F., Monteiro, M., Andrade, J., Bramati, I.E., Vianna-Barbosa, R., et al., 2014.Structural and functional brain rewiring clarifies preserved interhemispheric transferin humans born without the corpus callosum. Proc. Natl. Acad. Sci. U. S. A. 111,7843–7848.

Utsunomiya, H., Yamashita, S., Takano, K., Okazaki, M., 2006. Arrangement of fiber tractsforming probst bundle in complete callosal agenesis: report of two cases with anevaluation by diffusion tensor tractography. Acta Radiol. 47, 1063–1066.

Vergani, P., Ghidini, A., Strobelt, N., Locatelli, A., Mariani, S., et al., 1994. Prognostic indica-tors in the prenatal diagnosis of agenesis of corpus callosum. Am. J. Obstet. Gynecol.170, 753–758.

Wahl, M., Strominger, Z., Jeremy, R.J., Barkovich, A.J., Wakahiro, M., et al., 2009. Variabilityof homotopic and heterotopic callosal connectivity in partial agenesis of the corpuscallosum: a 3 T diffusion tensor imaging and Q-ball tractography study. AJNR Am.J. Neuroradiol. 30, 282–289.

Watts, D.J., Strogatz, S.H., 1998. Collective dynamics of ‘small-world’ networks. Nature393, 440–442.

Xia, M., Wang, J., He, Y., 2013. BrainNet viewer: a network visualization tool for humanbrain connectomics. PLoS ONE 8, e68910.

Zalesky, A., Fornito, A., Bullmore, E.T., 2010. Network-based statistic: identifying differ-ences in brain networks. Neuroimage 53, 1197–1207.