Decreased and Increased Anisotropy along Major Cerebral White Matter Tracts in Preterm Children and Adolescents

Premature birth is highly prevalent and associated with neurodevelopmental delays and disorders. Adverse outcomes, particularly in children born before 32 weeks of gestation, have been attributed in large part to white matter injuries, often found in periventricular regions using conventional imaging. To date, tractography studies of white matter pathways in children and adolescents born preterm have evaluated only a limited number of tracts simultaneously. The current study compares diffusion properties along 18 major cerebral white matter pathways in children and adolescents born preterm (n = 27) and full term (n = 19), using diffusion magnetic resonance imaging and tractography. We found that compared to the full term group, the preterm group had significantly decreased FA in segments of the bilateral uncinate fasciculus and anterior segments of the right inferior fronto-occipital fasciculus. Additionally, the preterm group had significantly increased FA in segments of the right and left anterior thalamic radiations, posterior segments of the right inferior fronto-occipital fasciculus, and the right and left inferior longitudinal fasciculus. Increased FA in the preterm group was generally associated with decreased radial diffusivity. These findings indicate that prematurity-related white matter differences in later childhood and adolescence do not affect all tracts in the periventricular zone and can involve both decreased and increased FA. Differences in the patterns of radial diffusivity and axial diffusivity suggest that the tissue properties underlying group FA differences may vary within and across white matter tracts. Distinctive diffusion properties may relate to variations in the timing of injury in the neonatal period, extent of white matter dysmaturity and/or compensatory processes in childhood.


Introduction
Preterm birth is a significant public health challenge, affecting approximately 12% of the infants born in the US [1]. Complications of preterm birth extend far beyond the newborn Within white matter, increased levels of FA have generally been associated with favorable neurobiological factors, such as increased myelination, greater axonal count and axonal density [21,22]. It has been argued that RD reflects myelin content and AD reflects axonal status [23][24][25][26][27], but such interpretations for diffusion properties may be oversimplified [22,28]. Other tissue factors may have counteracting effects on FA, such as directional coherence and the proportion of crossing-fibers within a voxel [29].
Some of the first dMRI studies to evaluate white matter status in children born preterm involved voxel-based analyses. At term equivalent age, these studies have generally found decreased FA and/or increased RD in preterm compared to the full term born neonates [6], consistent with reduced myelin content in the preterm samples. However, increased FA in specific regions has also been reported [30,31]. Differences in diffusion properties persist from the neonatal period into childhood [32][33][34], adolescence [35][36][37][38] and adulthood [39,40]. Studies of younger children have generally observed decreased FA in the preterm compared to full term sample [32][33][34]. Studies of adolescents and adults born preterm are inconsistent in their findings; studies find areas where preterm groups demonstrate significantly decreased FA [35,36], increased FA [38], both increased and decreased FA [37,39,40], or no differences between the preterm and full term groups [41]. A limitation of voxel-based approaches for group comparisons is that brain images must not only be initially co-registered to a common template based on brain shape and tissue boundaries, but must also be normalized or warped to ensure proper alignment of white matter tracts across subjects [28]. The requirements for normalization may be particularly problematic for comparisons involving clinical populations, such as children born preterm, where individual variability in ventricular size and/or unusual image contrasts are likely to be higher than in typical populations. Such variations can require higher degrees of warping to achieve normalization and may result in the comparison of non-equivalent white matter structures. Imprecise normalization opens the possibility that group differences may reflect partial volume effects rather than real differences in the properties of any specific white matter structure [28].
Diffusion MRI analyzed with tractography is generally regarded as the most precise method for fiber reconstruction within individual subjects [42]. Though prone to similar methodological limitations in terms of initial co-registration for anatomical referencing, tractography methods have been developed that do not require normalization for the comparison of equivalent white matter tracts across subjects [42,43]. Tracking within native space allows for the comparison of equivalent fiber pathways across individuals even if those tracts are in slightly different locations within the brain. Tractography methods have increasingly been used to identify and interrogate white matter status of specific fiber pathways in children born preterm. The dMRI tractography studies that compare preterm neonates at-and near-term equivalent ages with full term neonates provide evidence of early impaired connectivity [44][45][46]. Tractography studies also find associations between perinatal diffusion measures and clinical outcomes [7,16,44,[47][48][49][50][51][52] and FA increases with increasing gestational age [51]. Group differences in diffusion properties of white matter in preterm compared to full term samples have also been investigated in older children [9,10,12] and adolescents [11,13]. These structural studies have generally observed decreased mean FA of white matter tracts in the preterm sample compared to the full term sample [9][10][11][12]. However, studies have also observed regions of increased FA in the preterm sample [11] or no difference between the groups [13].
Tractography studies of preterm samples have typically focused on analysis of a single fiber tract or a small group of functionally related pathways, possibly because manual tractography approaches are laborious and time-consuming. Recently developed automated approaches for fiber tracking [42,53] now afford the ability to rapidly identify multiple white matter tracts within individuals of a given sample. An advantage of the Automated Fiber Quantification method (AFQ) [42] is that it identifies equivalent brain structures in native space within each subject individually, and also allows for the comparison of diffusion properties along the trajectory of each tract (tract profiles) rather than a comparison of averaged diffusivity values across the entire tract. Analysis of tract profiles has the advantage of revealing localized group differences that might otherwise be obscured by entire tract means, while also limiting the possibility of including tract regions where partial voluming with gray matter may be more prevalent [11,42,54,55]. In the current study, we compared tract profiles from a sample of children and adolescents born either preterm or full term to determine if group differences would occur throughout periventricular tracts segments and also diffusely in tracts non-adjacent to the lateral ventricles. In addition, we compared tract profiles of the preterm and full term groups to determine whether preterm children would demonstrate increased or decreased FA both within and across different tract trajectories in comparison to full term controls. Based on the evidence described above, we hypothesized that we would observe group differences in the tracts known to course adjacent to the lateral ventricles: the cortical-spinal tracts, forceps major, forceps minor, inferior fronto-occipital fasciculus, and inferior longitudinal fasciculus [56]. We anticipated that the preterm group would demonstrate both decreased and increased FA compared to the full term group within these tracts, based on previous findings in a voxel-based analysis of this cohort [38] and on recent tractography results in children and adolescents born preterm [11]. We explored evidence for diffuse white matter impairments by examining evidence for group differences in FA within tracts that do not course adjacent to the lateral ventricles: the arcuate fasciculus, uncinate fasciculus, anterior thalamic radiations, cingulum and anterior superior longitudinal fasciculus. In general, group differences identified in this age range were expected to represent the long-term consequences of white matter injury and/or dysmaturity that may occur after preterm birth.

Materials and Methods Participants
Demographic data for the preterm and full term groups is presented in Table 1. The groups did not differ significantly in age, gender, handedness, maternal education, and non-verbal or verbal IQ. [38,[57][58][59][60][61]. Maternal education level was dichotomized with "low" defined as less than a college degree and "high" as at least a college degree. IQ was measured by the Wechsler Abbreviated Scale of Intelligence (WASI), a widely used, nationally standardized test of general intellectual ability [63]. By design, the preterm sample had significantly decreased gestational age and birth weight than the full term sample. One full term and three preterm subjects were left-handed, and three full term and two preterm subjects were ambidextrous, as measured by the Edinburgh Handedness Inventory [64]. Review of T1 images revealed that 4 of the 27 preterm subjects had extreme ventricular enlargement that might affect measurement of diffusion properties in the periventricular regions [55]. The analyses described below were repeated excluding these subjects to assure that the reported group differences were not driven by their data.
Medical information was available for 25 of 27 preterm subjects. Medical complications at birth in the preterm group were: six with abnormal findings on head ultrasounds or MRIs ( grade 2 intraventricular hemorrhage, echodensities, or cystic lesions), two with mildly abnormal findings (grade 1 hemorrhage or choroid plexus cyst); 12 had respiratory distress syndrome, five developed bronchopulmonary dysplasia (BPD) or chronic lung disease; four had patent ductus arteriosus (PDA); none had necrotizing enterocolitis (NEC); one was small for gestational age ( 3rd percentile birth weight for gestational age).
A neuroradiologist unaware of the child's medical conditions evaluated the T1-and T2-weighted MRI scans of the preterm group, using a white matter scoring system derived from [65]. Each scan was scored from 5 to 15, based on a 3-point scale for five white matter features: white matter signal abnormality, periventricular white matter volume loss, cystic abnormalities, ventricular dilation, and thinning of the corpus callosum. We classified the scans as normal (score 6), mild-moderate (>6 and 12), or severe (>12) injury. Of the 27 preterm subjects, 13 were classified as normal, 10 were classified as having mild-moderate injury, and 4 as having severe injury. Previous evidence has shown that dMRI can be sensitive to microstructural differences in white matter in the absence of gross structural injuries in preterm neonates [66] and adolescents born preterm [11]. Moreover, a study of white matter volumes found that the results of both early and concurrent MRI scans did not affect white matter measurements [67]. We included all preterm subjects in group analyses regardless of gross white matter injury scores, consistent with other studies of this population [11]. However, we analyzed the data with and without the children with severe injury included in the group (see below). .9375 x 0.9 mm, TI = 300 ms, flip angle = 15 degrees, NEX = 1). T2 fluid attenuated inversion recovery (FLAIR) images were also collected (repetition time 9500ms, echo time 120ms, FOV 220, slices 5.0mm, no gaps) but were not analyzed. For dMRI and tractography, a diffusion-weighted, single-shot, spin-echo, echo-planar imaging sequence (TE = 80 ms, TR = 6500 ms, FOV = 240 mm, matrix size = 128 x 128, voxel size = 1.875 x 1.875 x 2 mm) was used to acquire 60 slices, in 30 different diffusion directions (b = 900). The sequence was repeated 4 times for improved SNR, and 10 non-diffusion weighted (b = 0) volumes were collected as well.

Data Preprocessing
The T1 images were co-registered to each other using a mutual information maximization algorithm (SPM5, http://www.fil.ion.ucl.ac.uk/spm/) and subsequently averaged for improved contrast. A trained research assistant manually identified the anterior and posterior commissures on the midsagittal plane, and these points were used to align the averaged anatomical image to a canonical ac-pc orientation, using a rigid body transformation (implemented in SPM5; no warping was applied).
Diffusion MR images were pre-processed with open-source software, mrDiffusion (http:// white.stanford.edu/newlm/index.php/MrDiffusion) implemented in MATLAB R2012a (Mathworks, Natick, MA). Eddy current distortions and subject motion in the diffusion weighted images were corrected by a 14-parameter constrained non-linear co-registration algorithm based on the expected pattern of eddy-current distortions, given the phase-encoding direction of the acquired data [68]. No subjects were excluded due to excessive head motion and no volumes were discarded. Additional analyses confirmed that full term and preterm groups also did not significantly differ in terms of translational head motion detected during image preprocessing (see S1 Text). Diffusion data were aligned to the T1 anatomical scans that had been averaged and rotated to align with the ac-pc plane. Alignment between dMRI and T1 data was achieved by registering the b0 images to the resampled T1 image using the same mutual information maximization algorithm used for T1 image coregistration provided through SPM5.
For each voxel in the aligned and resampled volume, tensors were fit to the diffusion measurements using a robust least-squares algorithm designed to remove outliers and data points corrupted by motion at the tensor estimation step (Robust Estimation of Tensors by Outlier Rejection (RESTORE) [69]. A continuous tensor field was estimated using trilinear interpolation of the tensor elements. We computed the eigenvalue decomposition of the diffusion tensor and the resulting three eigenvalues (λ1, λ2, λ3) were used to compute fractional anisotropy (FA), axial diffusivity (AD), and radial diffusivity (RD) [19]. The accuracy of image co-registration was verified in individual subjects by overlaying a white matter mask that included diffusion image voxels with FA >0.2 onto the average T1 image and visually inspecting the anatomical alignment.
AFQ consists of three major processing steps: (1) whole-brain tractography (2) automatic tract segmentation and cleaning, and (3) fiber quantification [42]. Whole brain tractography was performed using a deterministic streamlines tracking algorithm (STT) [69][70][71], with a fourth-order Runge-Kutta path integration method [72]. The fiber tracking algorithm was seeded with a white matter mask defined as all the voxels with FA value greater than 0.2 in the entire brain volume. Tracking proceeded in all directions and stopped when FA dropped below 0.15 or when the angle between the extension of a line in the direction of the current step and the direction of the next step was greater than 30°.
Tract segmentation was achieved using a multiple waypoint ROI procedure as defined by [73] and automated in AFQ [42]. During AFQ processing, an estimated non-linear transformation [74] was applied to automatically warp predefined ROIs from the Montreal Neurological Institute (MNI) template into an individual's native space. In this approach, ROIs are defined such that they isolate the central portion of the tract where fibers are most coherently bundled. Fibers are considered to belong to a specific tract only if they pass through both waypoint ROIs as specified in [73]. Using this procedure, we isolated for each individual 18 major pathways in the participant's native space. This included the posterior and anterior segments of the corpus callosum (forceps major, FMajor and forceps minor, FMinor, respectively), and 8 pairs of bilateral pathways: the anterior thalamic radiations (ATR), corticospinal tract (CST), cingulum (Cing), inferior fronto-occipital fasciculus (IFOF), inferior longitudinal fasciculus (ILF), anterior superior longitudinal fasciculus (aSLF), uncinate fasciculus (UF) and the arcuate fasciculus (Arc). Tracts were cleaned automatically using a statistical outlier rejection algorithm [69] for removing outlier fibers.
In a small proportion of tracts and individuals, cleaned tracts still included looping fibers that re-crossed both ROIs: Cing-L (4 preterm subjects; 1 of 4 had enlarged ventricles); FMajor (1 preterm subject; 0 of 1 had enlarged ventricles); aSLF-L (1 full term subject). These fibers were removed using Quench, a gesture based segmentation and visualization tool (http://white. stanford.edu/newlm/index.php/QUENCH). No other tracts required manual editing. The same investigator (KET) also verified that ROI placement for fiber segmentation was anatomically accurate in the small proportion of subjects in whom tracts were not reliably identified (see S1 Table). This procedure ensured that failure to detect these tracts was unlikely to be a consequence of inaccurate ROI placement.

Fiber Tract Quantification
FA was calculated at 30 equidistant nodes along a central portion of each fiber tract bounded by the same two ROIs used for tract segmentation. Tract extremities beyond these ROIs were not included in the analysis. This procedure generates, for every tract and every individual, an FA tract profile that describes the variations in FA along the central portion of the tract. At each node, diffusion properties (FA, RD, AD, MD) were calculated by taking a weighted average across all fibers belonging to this tract. Each fiber's contribution to the average was weighted by the probability that a fiber was a member of the fascicle, computed as the Mahalanobis distance from the tract core [42]. This procedure minimizes the contribution of fibers located further from the fiber tract core that are more likely to reflect a mixture of gray and white matter or of different tracts, and so minimizes the effect of partial voluming on diffusion estimates.

Classification of periventricular tracts and nodes
Tracts were first classified as periventricular or non-periventricular based on a white matter atlas [56]. A tract was considered to be periventricular if its trajectory traverses adjacent to the lateral ventricles. Based on this definition, we classified the CST, FMajor, FMinor, IFOF, and ILF as periventricular tracts and the Arc, UF, ATR, Cing and aSLF as non-periventricular tracts. We next loaded tract renderings of each of the periventricular tracts from representative full term and preterm subjects in Quench, to visually approximate the location of the lateral ventricles and to confirm that the ROI bounded portion of the tract traverses adjacent to the lateral ventricles. Nodes that were adjacent to the lateral ventricles were approximated on tract renderings using ac-pc coordinate information provided for each node in the tract profile.

Volume estimation
Volume estimates were obtained for the core region of each tract (ie., region between ROI1 and ROI2) and the whole-brain fiber group for each subject. Volume was estimated as the number of voxels covered by one or more streamlines in the aligned T1-image [75]. Volume estimates for each tract were normalized by dividing each tract volume estimate by the volume estimate for the whole brain fiber group. This step was performed to confirm that differences in tract diffusion properties did not reflect a general difference in white matter volume between full term and preterm groups.

Statistical Analyses
Analyses of demographic variables. Chi-square tests and two-tailed t-tests for independent samples were used to examine differences between the preterm and full term samples on demographic variables.
Group comparisons of tract properties: omnibus tests. Given that white matter changes are known to occur during the age range of the current sample [76], we first assessed FA in relation to age in the preterm and full term groups separately to determine whether it was necessary to control for age in group analyses. This step was achieved by calculating Spearman correlations between age at DTI scan and mean FA for each tract. In the full term group, we found three tracts (Arc-R, Cing-L, Cing-R) in which age associations were present. In the preterm group, we found an age association within the ILF-L. The resulting statistics of this analysis are presented in S2 Table. To identify age-associations that may have been obscured by whole tract mean measures, we also calculated Spearman correlations between age at DTI scan and FA at 30 locations along each tract. These analyses revealed additional significant positive age-associations (p < 0.05, uncorrected) within the full term (CST-L, CST-R, ATR-L, ATR-R,) and preterm groups (CST-R, FMinor, UF-R, IFOF-R) that were not apparent in analyses with whole tract mean measures. Given these findings, we included participant's age at the time of scanning in the subsequent ANCOVAs as a continuous covariate in order to isolate group differences independent of age. For consistency, we include age at DTI scanning as a continuous covariate for all tracts.
We assessed group differences in tract properties by calculating, for each tract, a mixed design three-way analysis of co-variance (ANCOVA) [77], with Group (preterm versus full term) as a between-subject factor, while Hemisphere (left versus right tract homologs) and Location (30 nodes along the tract) served as within-subject measures. FA profiles of the Forceps Major and Forceps Minor were divided along the midsagittal plane to allow a comparison of their left (nodes 1-15) and right (nodes [16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] sections, similarly to the left and right homologs of each tract. Degrees of freedom were corrected using Greenhouse-Geisser estimates in cases where sphericity was violated [78]. Group differences were considered to be significant at p < 0.05. To control for multiple comparisons across tracts, the significance of omnibus tests was determined using a false discovery rate (FDR) of 5% [79].
To ensure that the results of this analysis were not driven by the four preterm subjects with extreme ventricular enlargement, ANCOVA analyses were repeated without these subjects. To determine if the present pattern for group differences in FA were secondary to differences in volume, we calculated t-tests for independent samples to examine whether mean volume for all tracts and the whole brain fiber group systematically varied between preterm and full term groups. T-tests for independent samples were then repeated using normalized tract volume for those tracts found to exhibit significant group differences.
Finally, since previous studies of older preterm children and adolescents have reported group differences in MD [11,40], we also repeated ANCOVA analyses with mean diffusivity (MD) for completeness. Spearman correlations calculated between age at DTI scan and mean tract MD revealed significant age associations (p<0.05) in 9 out of 18 tracts in the full term group (S3 Table). In the preterm group a significant age-association was observed only for the IFOF-R (S3 Table). Based on these findings, we included age at DTI scan as a continuous covariate to isolate group differences with MD that were independent of age. For consistency, we included age at DTI scanning as a continuous covariate for all tracts. The significance of omnibus tests with measures of MD was determined using a false discovery rate (FDR) of 5% [79].
Location-and hemisphere-specific contrasts. To identify the hemisphere and tract segments (nodes) responsible for the group differences identified by the omnibus tests, we followed up on significant ANCOVA effects with sets of two-tailed t-tests for independent samples, comparing the FA values of the preterm and full term groups at each node along the tract profile (separated by hemisphere). These comparisons were calculated only for tracts identified by the ANCOVAs such that: (1) the tract showed a significant main effect of Group, or (2) the tract showed a significant interaction between Group and Location, or (3) the tract showed a significant interaction between Group and Hemisphere. We considered (2) and (3) essential because group effects could have occurred in opposing directions within different segments of the same tract or in different hemispheric homologs of the same tract. Location and hemisphere specific contrasts were not investigated further if group differences did not survive correction for multiple comparisons across tracts.
We employed a nonparametric permutation-based method to control for the 30 comparisons along the tract [80]. This procedure produced a family-wise error corrected cluster size and a critical t-value for each of the candidate tracts. Tract segments were considered significant if differences occurred either (1) in a sufficient number of adjacent nodes to meet the criteria for a family-wise error corrected cluster size or (2) in nodes in which the effect size was greater than the critical t-value.
Exploratory analyses: gestational age, AD and RD. To further interrogate group differences, we analyzed the association between individual gestational age at birth in the preterm group and FA within tract segments that demonstrated significant group differences. We calculated Spearman correlations between gestational age at birth and mean FA extracted from the segments where significant group differences were detected in the along tract profile comparisons. These analyses were limited to the preterm group, to avoid a pseudo-correlation reflecting the group differences in both variables. To investigate the contributions of AD and RD to group differences in FA values, we computed a separate one-way multivariate analysis of variance (MANOVA) for each cluster of nodes identified in the tract profile analysis as showing a significant group difference in FA. Group (preterm versus full term) served as the between-subject variable and mean AD and mean RD served as the dependent variables. We chose to enter AD and RD into the same model in order to reduce the number of comparisons and increase power for detecting group differences.

Fiber Tracking
Using automated fiber segmentation procedures, we successfully identified 18 cerebral tracts of interest in the majority of both full term and preterm individuals (Fig 1). A small number of tracts were missed in one to three subjects (see S1 Table). In addition, we were unable to identify the Arc-R in 6 full term and 4 preterm subjects. We attribute the difficultly in identifying the Arc-R to limitations of deterministic tractography approaches that cannot account for higher tract curvature and increased partial voluming with the aSLF-R, a finding consistent with several other reports [43,[81][82][83].  Fig 1B and 1D. Visual inspection of the suspected periventricular tracts (CST, FMajor, FMinor, IFOF and ILF) in representative full term and preterm subjects confirmed that these tracts contained regions adjacent to the lateral ventricles. These periventricular regions are marked by light shading on the relevant tract renderings in Figs 2 and 3. Tract results are first described based on the order in which they appear in Figs 2 and 3 ( Table 2). For comparative purposes, group mean tract FA values for all 18 cerebral tracts are presented in Table 3. In subsequent sections, this same ordering is then sub-divided by the direction of group differences: Preterm < Full term and Preterm > Full term (Tables 4 and 5).

Group comparisons of tract properties: Omnibus Tests (FA)
ANCOVAs calculated using measures for FA and age as a covariate for each tract revealed significant main effects of Group in the ATR and ILF (Table 2). Significant interactions were found between Group and Location in the FMajor, UF, IFOF, and ILF. Group comparisons remained significant after controlling for multiple comparisons across tracts at a 5% criterion for FDR for all tracts except for the FMajor. For this reason, location and hemisphere specific contrasts were not investigated further for the FMajor. No tracts demonstrated significant Group by Hemisphere interaction effects. No significant group effects or interactions were observed for the following five tracts: Arc, CST, FMinor, Cing and aSLF. In addition, no tracts demonstrated significant main effects or interactions with age except for a significant Age by Location interaction effect in the cingulum F(4.91,206.3) = 4.02 p < 0.002.
Post-hoc analyses confirmed that group effects identified in the omnibus tests remained significant (p < 0.05) after removing the four preterm subjects diagnosed with extreme ventricular enlargement. Since omnibus effects did not appear to be driven by these 4 cases, these subjects were included in all subsequent analyses.
Analyses of tract volume confirmed that group differences in FA were unlikely to be driven by systematic differences in tract volumes or volume of the whole brain fiber group (S4 Table). Tract volume was lower for the preterm than full term group in the FMajor and IFOF-R but no significant tract volume group differences were observed for any of the other 16 cerebral white matter tracts examined or for the whole brain fiber group from which individual tracts had been segmented (S4 Table). These group differences within the FMajor and IFOF-R remained significant (p < 0.05) after controlling for the size of the whole brain fiber group. Further, the full term group was found to have significantly higher tract volume in tracts where the full term group demonstrated both significantly increased and decreased FA compared on the preterm group (S4 Table). Specifically, within the FMajor, the full term group demonstrated significantly (p < 0.05) higher tract volume than the preterm group. In the IFOF-R, the full term group demonstrated significantly (p < 0.05) higher tract volume than the preterm group, though the full term group was observed to demonstrate regions of both higher and lower FA.

Group comparisons of tract properties: Omnibus Tests (MD)
ANCOVAs calculated using measures for MD and age as a covariate revealed a significant main effect of Group in the ILF (S5 Table). Significant interactions were found between Group and Location in the FMajor and FMinor (S5 Table). No tracts demonstrated significant Group by Hemisphere interaction effects. A significant main effect of age was observed for the UF (F   (2.55,109.84) = 2.96, p = 0.04). Group comparisons were no longer significant after controlling for multiple comparisons across tracts at a 5% criterion for FDR. For this reason, location and hemisphere specific contrasts were not investigated further.

Location-and Hemisphere-specific contrasts (FA)
In each tract that showed a significant main effect of Group or a significant interaction with Group, we compared the mean FA values for the two groups at each node along the tract, using two-tailed t-tests. Tract segments showing significant group differences are indicated in Figs 2 and 3 with a light gray background shading (preterm < full term) or a light brown background shading (preterm > full term). The pattern of results was generally similar in the left and right hemispheres, though the differences did not always rise to the level of statistical significance in both hemispheres. FA was significantly decreased in the preterm group compared to the full term group in the UF bilaterally (Fig 2G and 2H) and anterior segments of the IFOF-R ( Fig  3F); see Table 4 for detailed statistics. In contrast, FA was significantly increased in the preterm group compared to the full term group within the ATR bilaterally (Fig 3A and 3B), posterior segments of the IFOF-R (Fig 3F), and the ILF bilaterally (Fig 3G and 3H) (Table 4). In general, these detailed contrasts revealed that the preterm group demonstrated significant decreases as well as significant increases in FA relative to the full term group, in both periventricular and non-periventricular tracts (Figs 2 and 3). Moreover, within specific periventricular tracts left (g) and right (h) inferior longitudinal fasciculus; and left (i) and right (j) anterior superior longitudinal fasciculus. Dashed lines indicate ± 1 standard error of the mean. FA values are plotted for 30 equidistant nodes between two ROIs used to isolate the core of each tract. For all tracts Node 0 corresponds to ROI-1 and Node 30 corresponds to ROI-2. Tracts with significant main effect of Group in the omnibus test are indicated with a red *. Tracts with significant Group by Location interactions in the omnibus test are indicated with a blue *. Shaded gray regions indicate tract segments where mean FA of the PT group is significantly decreased than the FT group. Shaded light brown regions indicate tract segments where mean FA of the FT group is significantly increased than the PT group. Tract renderings from a representative full term subject pictured in Fig 1A and 1B

Associations with Gestational Age within the Preterm group
We performed Spearman correlations to determine if the gestational age at birth of the preterm sample was associated with FA within tract segments that demonstrated significant group differences. Correlation coefficients for each tract segment are reported in Table 4. This analysis revealed a significant positive association between gestational age and FA in an anterior

Group Comparisons: AD and RD
In tract segments in which we detected a significant group difference in FA, we conducted MANOVAs in order to assess whether this difference would be best explained by differences in AD or RD. These analyses revealed significant multivariate effects of group (FT vs PT) for all tracts (p < 0.05). To identify whether group effects identified in the MANOVA were the result of group differences in AD, RD or both, we next performed post-hoc univariate ANCOVAs. The results of this analysis are reported in Table 5. In the three tract segments in which the preterm group had decreased FA compared to the full term group (UF-L, UF-R, and IFOF-R), the preterm group showed increased AD compared to the full term group. In tract segments in which FA was significantly increased in the preterm group compared to the full term group, the preterm group had significantly decreased RD compared to the full term group in the following tracts: ATR-L, ATR-R, IFOF-R, ILF-L, and ILF-R. The preterm group also had significantly increased AD compared to the full term group, with both AD and RD contributing to increased FA in the IFOF-R and the ILF-R.

Summary of Results
Using an automated, large-scale tractography approach, we identified 18 cerebral white matter tracts in a sample of children and adolescents born preterm and full term. We found significant   group differences in FA within multiple tracts. Compared to the full term controls, the group of children born preterm had decreased FA in the uncinate fasciculus bilaterally and an anterior segment of the right inferior fronto-occipital fasciculus. By contrast, the preterm group had increased FA within the anterior thalamic radiations, posterior segments of the right inferior fronto-occipital fasciculus, and the inferior longitudinal fasciculus. These group differences occurred both within periventricular and non-periventricular tracts segments. Group differences did not appear to be explained by the 4 preterm subjects with ventricular dilation or by individual variations in tract volume. With the exception of the forceps minor, ANCOVA analyses repeated using MD did not reveal evidence for further microstructural differences in tracts beyond those identified using FA. Exploratory analyses revealed a significant positive association between FA and gestational age at birth of the preterm born individuals within an anterior region of the right inferior fronto-occipital fasciculus. Increased FA in the preterm group was generally associated with decreased RD, while decreased FA in the preterm group was associated with increased AD.

Group differences in periventricular tracts
Consistent with our initial predictions, FA differences between the preterm and full term groups were found within periventricular white matter tracts, including the inferior frontooccipital fasciculus and the inferior longitudinal fasciculus. Specific contrasts revealed that group differences observed in the right inferior longitudinal fasciculus were found specifically in the periventricular zones within these tracts. These findings are generally consistent with the pathological characteristics of preterm brain injury [16]. They are also consistent with evidence from animal models of prematurity that document the vulnerability of periventricular white matter to injury from hypoxic-ischemic and inflammatory responses experienced during late gestation [84,85]. In contrast, group differences observed in the right inferior fronto-occipital fasciculus and left inferior longitudinal fasciculus localized to segments that were not directly adjacent to the lateral ventricles. More research is needed to determine whether group differences observed in regions distal to the lateral ventricles reflect white matter injuries or dsymaturation arising from similar or distinct neuropathological processes from those observed in periventricular areas. Such studies may benefit from the use of animal models of prematurity [86] or MR spectroscopy for distinguishing amongst cellular mechanisms related to axonal dysmaturity or astrogliosis that may contribute to white matter injuries associated with prematurity [87]. Contrary to the initial hypothesis, FA of the periventricular left or right corticospinal tracts and forceps minor did not differ significantly between preterm and full term groups. There are several possible interpretations for the lack of differences within these periventricular tracts. One possibility is that injuries in these pathways observed in other studies at term equivalent ages [6,44,47] may have resolved by the ages studied here. Age-related increases in FA are well documented in children [88][89][90] and could thus compensate for initial decreases in FA. A second possibility is that the differences resolved because the present sample of preterm children received enhanced education and therapy. White matter microstructural changes are demonstrated to occur in response to a range of environmental experiences in numerous animal species [91] and in response to intensive behavioral remediation in humans [92][93][94][95][96]. Early injury to the forceps minor and corticospinal tracts [6,44,47] could thus recover as children practice the visual or motor skills dependent on axonal projections that comprise these tracts.
The lack of significant group differences within corticospinal tracts and the forceps minor may also be explained by several methodological differences across studies. The procedures we used for identifying and segmenting tracts were designed to characterize core tract regions.
Our procedures purposely excluded peripheral tract segments that are more variable and susceptible to partial voluming [42]. The present methods may thus be insensitive to group differences, that in previous voxel-based studies, have been shown to occur in regions of the corticospinal tract that extend beyond the core tract regions analyzed here [6,38]. By ignoring peripheral tract regions, the present methods may also be insensitive to group differences that may occur in more distal regions that can otherwise be captured by mean tract measures generated from averaging diffusion properties over the entire length of white matter pathways [10]. Alternatively, it is possible that by isolating only core tract regions, the present methods capture fibers that may be robust to injury and dysmaturation or have enhanced capabilities for recovery. Finally, it is also possible that the present pattern of observations relates to specific clinical characteristics of the present sample. Investigating these possibilities will ultimately require replication in other samples, as well as longitudinal analyses to understand whether FA differences observed at younger ages change in association with age or specific environmental experiences.

Group differences in non-periventricular tracts
In the present study, we observed evidence for microstructural differences between the preterm and full term groups in non-periventricular tracts, including the anterior thalamic radiations and the uncinate fasciculus. These findings are consistent with previous tractography studies that have found group FA differences within the left and right anterior thalamic radiations in preterm neonates with varying degrees of white matter abnormalities [97], as well as findings from a tractography study in which a group of preterm children demonstrated reduced FA in both the left and right uncinate compared to the full term group [9]. Together with these findings, the present pattern of results suggests that white matter changes associated with preterm birth are also diffuse [4,16].
Evidence for microstructural differences within the anterior thalamic radiations and uncinate fasciculus also have important implications for understanding the range of neurodevelopmental outcomes experienced by children born preterm. Axonal projections contained within the anterior thalamic radiations are known to connect the thalamus, striatum, and frontal lobes [98]. The uncinate fasciculus connects regions of the anterior temporal and inferior frontal lobes and has been implicated in higher-order cognitive and linguistic skills [99]. Evidence for injury or dysmaturity to fiber connections of the anterior thalamic radiations in preterm children may account, in part, for individual variations in behavioral abilities thought to involve functions of the frontal lobes, including attention [100][101][102] and working memory processes [103][104][105]. Similarly, evidence for injury or dysmaturity within the uncinate fasciculus may account for the range of language and reading outcomes observed in preterm children [106][107][108][109].
The present study cannot determine whether the group differences observed in the anterior thalamic radiations or uncinate fasciculus reflect the long-term consequences of early injury or altered maturation, or differences related to developmental experiences after birth. Multiple sources of evidence indicate that myelination of supramodal frontal and temporal cortices has a protracted developmental time-course relative to unimodal cortices responsible for supporting primary sensorimotor processes [110,111]. Longitudinal studies will be important for determining whether the observed group differences seen here and in other studies of older preterm children and adults [9][10][11][12] reflect sustained injuries from birth or delayed differences reflective of altered maturation in the myelination of these pathways or compensatory processes. In addition, intervention studies will be important for determining whether microstructural properties of white matter tracts are capable of changing in association with specific educational or medical therapies for improving functional outcomes. Such findings would have important implications for understanding neural basis of plasticity following early white matter injury and functional characteristics of white matter pathways in general.

Gestational Age
In this study we observed very few statistically significant associations between FA and gestational age at birth of the preterm participants. Studies find that behavioral outcomes of children born preterm tend to correlate with gestational age at birth, including cerebral palsy, which itself is associated with brain injury [112]. Studies have also found that selected brain findings, such as bi-parietal width, are also correlated with gestational age at birth [113]. Such relationships suggest that white matter properties associated with poor outcomes might also be associated with gestational age. Indeed, studies have observed associations between white matter properties in relation to gestational age at birth [45,49,[114][115][116][117][118][119]. However, it has also been shown that accumulated medical variables over the course of the neonatal hospitalization are more predictive than gestational age [120,121]. Several additional explanations may account for the minimal significant correlations observed here. First, sample sizes must be adequately powered to observe associations. The present sample size may be too small to detect such correlations over and above individual differences within the preterm group. Second, correlations between diffusion properties and gestational age may be strongest at the limits of viability, similar to correlations with behavioral outcomes [122]. Here, the gestational age range of the participants in this study was broad. Third, white matter injuries may be associated with the timing of adverse complications from preterm birth, such as hypoxia-ischemia and infection, and these events may occur pre-, peri-or post-natally. Diffusion properties may therefore be more closely associated with the presence of acute adverse medical events than with gestational age per se [47]. Samples larger than the present one would be required to investigate such associations and will be useful for performing sub-group analyses to determine whether group differences on the basis of preterm birth may be driven by specific gestational age-ranges.

Variations in the diffusion properties across tracts
Across the different tracts assessed in this study, we found that the preterm group had both decreased and increased FA relative to the full term group. Such findings suggest that the underlying neurobiological factors that contribute to FA differences vary across and along the trajectory of individual white matter tracts.
Evidence for decreased FA within anterior segments of the right inferior fronto-occipital fasciculus and bilateral uncinate fasciculus in the preterm group is consistent with several previous studies of preterm children and adults at various ages [9][10][11][12]. Interestingly, decreased FA was not detected within this same sample using a different diffusion tensor imaging analytic strategy [38]. The different results using the two methods suggest greater sensitivity of tractography to group differences in this clinical group. The primary assumption about decreased FA in children born preterm is that the results suggest decreased myelination. Groeschel and colleagues (2014) specifically postulated that decreased FA in the preterm group would occur in segments of the corpus callosum and internal capsule where there was a single fiber orientation. In these regions, increased radial diffusivity appeared to be the largest contributor to decreased FA [11]. Our results were quite similar for the anterior segments of the right inferior frontooccipital fasciculus where we too found decreased FA and increased radial diffusivity. However, within the uncinate fasciculus bilaterally and also within anterior segments of the right inferior fronto-occipital fasciculus, the preterm group in the current study had decreased FA and elevated AD. Any explanations for decreased FA in these tracts that we might offer would be speculative. Nonetheless, taken together, these variations in diffusion properties suggest that underlying tissue properties related to myelin and axonal status may vary and interact in different tracts with both changes leading to decreased FA.
Increased FA in several tracts of the preterm group is consistent with findings from this cohort using a different dMRI analytic strategy [38], as well as other dMRI studies of children, adolescents and adults born preterm [11,37,39,40]. Groeschel et al., (2014) located increased FA to predominantly crossing-fiber regions where the interpretation of radial and axial diffusivity is not meaningful. They attributed increased FA to a decrease in anisotropy of a contributing fiber bundle. A similar explanation would be a reduction in the number or density of crossing fibers from axonal loss in the preterm group that could result in increased FA [28]. Alternatively, increased FA may reflect increased myelination from compensatory processes. However, such interpretations for FA are only likely to reflect changes in myelin status within white matter tracts demonstrating single fiber orientations [22]. Studies have estimated that the majority of white matter voxels (60-90%) contain crossing fibers [29]. Taken together, such findings suggest that the observed patterns for both increased and decreased FA in the present study may reflect inter-and intra-tract variations in the proportions of crossing fibers. Distinguishing amongst these interpretations for FA is likely to require additional neuroimaging techniques for obtaining voxel-based estimates for the proportion of crossing fibers [123,124] and tissue myelination [125,126]. In general, combining such methods with FA measurements is likely to afford greater understandings for the neurobiological underpinnings of both decreased and increased FA observed in preterm children at different ages. For example, this combination may help to clarify whether increased FA within specific pathways is associated with high myelin content, or alternatively whether increased FA in the context of low myelin content may reflect decreased numbers of crossing fibers possibly related to axonal loss.

Limitations
The current group of children born preterm represented a convenience sample with a wide age range. Given the large age range of the current sample, it is possible that developmental changes in FA could have accounted for the present group differences. However, we think that this possibility was low because we did not observe significant main effects of age for any white matter tract in omnibus tests performed to identify group differences. Replication studies within specific age groups and/or larger sample sizes will be important for establishing the generalizability of the present findings.

Conclusions and Future Directions
This dMRI tractography study confirms that differences in white matter microstructure observed for children born preterm persist into adolescence. Group differences are found in both periventricular and non-periventricular white matter tracts. The pattern of results is complex: the preterm group shows both increased and decreased FA values in different segments of the identified white matter tracts compared to the full term group. Future studies using quantitative MRI measurements are needed to understand the sources of the identified differences. In addition, longitudinal analyses may help to determine how patterns of prematurity-related white matter differences change from birth to adulthood. Future studies involving larger samples of preterm and full term children than the current study will be important for understanding how other factors that are known to affect white matter development, such as sex and birth weight [77,127,128], interact with group differences. Finally, future studies should consider the functional implications of the structural differences identified here between children born preterm and full term.
Supporting Information S1 Fig. Spearman correlation between GA and mean FA from the right IFOF where significant group differences were observed (location [18][19]. (DOCX) S1