Frontal interhemispheric structural connectivity, attention, and executive function in children with perinatal stroke

Abstract Perinatal stroke affects ∼1 in 1000 births and concomitant cognitive impairments are common but poorly understood. Rates of Attention Deficit/Hyperactivity Disorder (ADHD) are increased 5–10× and executive dysfunction can be disabling. We used diffusion imaging to investigate whether stroke‐related differences in frontal white matter (WM) relate to cognitive impairments. Anterior forceps were isolated using tractography and sampled along the tract. Resulting metrics quantified frontal WM microstructure. Associations between WM metrics and parent ratings of ADHD symptoms (ADHD‐5 rating scale) and executive functioning (Behavior Rating Inventory of Executive Function (BRIEF)) were explored. Eighty‐three children were recruited (arterial ischemic stroke [AIS] n = 26; periventricular venous infarction [PVI] n = 26; controls n = 31). WM metrics were altered for stroke groups compared to controls. Along‐tract analyses showed differences in WM metrics in areas approximating the lesion as well as more remote differences at midline and in the nonlesioned hemisphere. WM metrics correlated with parental ratings of ADHD and executive function such that higher diffusivity values were associated with poorer function. These findings suggest that underlying microstructure of frontal white matter quantified via tractography may provide a relevant biomarker associated with cognition and behavior in children with perinatal stroke.

Neonatal arterial ischemic strokes are diagnosed soon after birth (following seizures), while arterial presumed perinatal ischemic strokes may be diagnosed later in infancy when early hand preference and hemiparesis become evident (Kirton & deVeber, 2013). These two AIS types only differ in timing of diagnosis (Dunbar & Kirton, 2019). By contrast, PVIs are venous infarctions resulting from a germinal matrix hemorrhage characterized by damage to the periventricular white matter. PVIs typically occur earlier than AIS, usually before 34-week gestation (Kirton et al., 2008) and damage is restricted to the deep periventricular white matter. Both AIS and PVI result in deficits and will likely continue to occur given a lack of identified modifiable risk factors or prevention strategies (Dunbar & Kirton, 2018). Given that these focal injuries occur so early in life, studying children after perinatal stroke may give unique insight into neuroplastic compensatory mechanisms unfolding during development.
How the nature of an early brain lesion interacts with developmental plasticity to produce variable outcomes in executive function and attention is not understood. Using MRI imaging techniques, differences in functionally connected neural networks of children with AIS compared to those with PVI and controls have been associated with cognitive function Ilves et al., 2016). Structural connectivity methods, including diffusion imaging (dMRI), allow for the isolation and characterization of specific white matter tracts in vivo. dMRI facilitates investigation of the diffusion of water across differing tissues within the brain and how larger molecules such as myelin, microtubules, and axons, restricts the movement of water (Basser, 1995;Beaulieu, 2002). Microstructural metrics such as fractional anisotropy (FA) and mean diffusivity (MD), among others, allow for quantification of the degree of diffusion in underlying tissue. Differences may reflect damage or differential development in white matter structures. These metrics as well as axial (AD) and radial diffusivity (RD) may provide additional information about developmental neuroplastic mechanisms following early brain injury. Our group and others have demonstrated the ability of tractography to better understand sensory and motor functions in children with perinatal stroke Kuczynski et al., 2017;Kuczynski et al., 2018;van der Aa et al., 2011). Furthermore, recent evidence has emphasized the essential role of widespread network alterations in the nonlesioned hemisphere in determining perinatal stroke outcomes (Craig et al., 2020), emphasizing the importance of interhemispheric connectivity.
Executive functions have largely been localized to areas of the frontal lobes using multimodal neuroimaging and subsequently, the importance of developing rich connectivity among frontal areas and other cortical, subcortical and limbic areas of the brain during maturation has been demonstrated (Fiske & Holmboe, 2019). Previous research using diffusion imaging metrics has shown an association between microstructural characteristics of frontal white matter, symptomology of ADHD, and executive function (Ashtari et al., 2005;Hong et al., 2014;Konrad & Eickhoff, 2010;Liston et al., 2011;Makris et al., 2009;Wu et al., 2020). Specifically, FA has been found to be lower in the frontal white matter of children with ADHD compared to peers (Konrad & Eickhoff, 2010;Liston et al., 2011;Tremblay et al., 2020;van Ewijk et al., 2012). Elevated MD in the anterior forceps of those with ADHD is a consistent finding across literature examining frontal tracts (Lawrence et al., 2013;van Ewijk et al., 2012). In adults, elevated MD was associated with longer reaction times during the Stroop task (a measure of cognitive interference), as well as decreased attention switching speed and flexibility (Mamiya et al., 2018). Findings related to other diffusivity metrics of the anterior forceps (radial [RD] and axial diffusivity [AD]) have been less consistent. AD tends to be higher in those with ADHD, but RD has shown mixed results (van Ewijk et al., 2012). In children with nonspecific unilateral cerebral palsy, differences in microstructure metrics in the anterior cingulate cortex have been associated with cognitive function (Scheck et al., 2015).
Given the high rates of ADHD and executive dysfunction in the perinatal stroke population, we investigated whether the character of frontal white matter in children with stroke would be altered compared to controls. We used dMRI tractography to examine the microstructure of the anterior forceps and possible relationships with parent ratings of ADHD and executive functioning. We hypothesized that the microstructure of frontal white matter would show disruptions (lower FA and higher MD, RD and AD) in children with AIS compared to children without injury (controls), and that children with PVI would be similar to controls given that cortical areas are typically preserved in this population. Further, we hypothesized that disruptions of frontal white matter (but not white matter in posterior lobes) would correlate with lower scores on executive function outcomes.

Participants
Children with perinatal stroke were recruited from a population-based research cohort, the Alberta Perinatal Stroke Project (APSP) (Cole et al., 2017). Inclusion criteria were (1) MRI-confirmed unilateral perinatal stroke following term birth (> 36 weeks) with no evidence of additional bilateral or diffuse injury as confirmed by a pediatric neurologist and (2)

Image acquisition
Imaging was performed at the Alberta Children's Hospital using a 3

Image processing
Anatomical T 1 W scans underwent segmentation into cerebrospinal fluid (CSF), gray, and white matter using Statistical Parametric Mapping (SPM12; Wellcome Centre for Human Neuroimaging, UCL London). An estimate of total intracranial volume was calculated by summing volumes of CSF, gray, and white matter. Segmentations were also used to generate a gray matter-white matter interface mask, which was subsequently used to restrict the generation of reconstructed streamlines to only white matter, termed anatomically constrained tractography . Anatomical scans and masks were linearly transformed into diffusion space using FSL's "FLIRT" followed by nonlinear transformation using "FNIRT" (Andersson et al., 2007;Jenkinson et al., 2012).

Regions of interest selection
Regions of interest (ROIs; Figure 1) were selected based on an anatomical model of the corpus callosum (Hofer & Frahm, 2006). A color map indicating directionality of water diffusion was overlaid on a T 1 W anatomical scan. In the axial view, the cursor was placed at the midline of the genu of the corpus collosum where the color map indicated primarily left-right diffusion direction. In the sagittal view, the hook of the genu was then traced and filled to isolate the anterior forceps ( Figure 1a).
The posterior forceps were also investigated for comparison and to assist in establishing functional specificity of subsets of the frontal white matter. A similar process was used to create the tracts for the posterior forceps, using an ROI placed over the splenium at the base of the posterior "bulb" of the corpus callosum at midline. Exclusion ROIs were drawn to demarcate spurious streamlines (typically the cingulum bundle) that were subsequently excluded during tract reconstruction.
For the anterior and posterior forceps separately, ROIs were used as seeds to select 5000 streamlines passing through them. This streamline threshold was chosen given previous findings of reliable white matter metrics using this degree of sampling (Reid et al., 2020). Resulting tracts were binarized using "tckmap" and then overlaid onto the tensor image where mean values of FA, MD, AD, and RD for the entire tract were extracted using "tensor2metric" and "mrstats" (Tournier et al., 2019).

Intrarater reliability
To assess the reliability of ROI placements and reconstruction of tracts, extraction of mean white matter metrics for the anterior forceps were repeated one month later by the same tractographer (NL) on a subset of 14 randomly selected participants (17% of the total sample).

Along-tract white matter metrics
To investigate possible diaschisis (degeneration of brain structures spatially displaced from the primary lesion), white matter metrics were The tckresample function in MRtrix3 calculated locations for 11 equidistant points along the reference arc and using a line perpendicular to the tangent of the arc, white matter metrics were extracted from these 11 samples. Resulting white matter metric values were reassigned for children with left-side stroke to match right-side stroke such that nonlesioned/dominant (segments 1-5), midline (segment 6), and lesioned/nondominant (segments 7-11) were compared appropriately among groups.

Lesion volume
Lesion volumes (in cubic centimeters [cc]) were measured using the 3-dimensional ROI selection tool in MRIcron (Rorden & Brett, 2000) based on T 1 W image intensity. For the AIS group, the center of the lesion was selected on an axial slice and the resulting dilated ROI was verified and adjusted manually (if necessary) on every axial slice. For the PVI group, bilateral ventricle volumes were measured as above and ventricle asymmetry (lesion size) was calculated as the absolute value of the difference between the two volumes as such: PVI lesion size = | lesioned hemisphere ventricle volume -nonlesioned ventricle volume |.

Cognitive function
Parents

Statistical analysis
The Statistical Package for Social Sciences version 28 (IBM SPSS, Armonk, New York) and R (R Core Team, 2017) were used for data analysis. All data were tested for normality using Shapiro-Wilk.
Demographic differences among the three participant groups were explored using Kruskal-Wallis (

Age correlations
Participant age was moderately correlated with some white matter ADHD ratings were not significantly correlated with age.

Differences in white matter metrics among participant groups
Anterior and posterior forceps were successfully reconstructed in all participants (Figure 1d-f). As illustrated in Figure 2, mean FA in the anterior forceps was different among groups (H (2) = 6.3, p = .043, ε 2 = .08, Table 2) such that FA was lower for AIS compared to controls (p = .048). Anterior forceps FA for children with PVI was not different compared to AIS or controls (p = .199 and p = 1.000, respectively).
For the posterior forceps, mean FA varied among groups (H (2) = 11.2, p = .004, ε 2 = .14). FA for the AIS group was significantly lower compared to controls (p = .002) and the FA of the PVI group fell in between AIS and controls, but was not different than either (p = .218 and p = .422, respectively, Figure 2b, Table 2).

White matter metrics and cognitive function
A subset of mean anterior forceps white matter metrics were associated with measures of cognitive function (Table 4, Figure 2e and f).
*p < .05 compared to controls. **p < .01 compared to controls. Total ADHD rating r s = .55, p = .014). A subset of these associations Note: Both the ADHD rating scale and the BRIEF are negatively scored such that higher numbers correspond to more symptoms of ADHD and executive dysfunction (i.e., poorer performance). r s , partial Spearman's correlation controlling for age. *p < .05. **p < .01. ***p < .0036 (Bonferroni correction).

Anterior forceps
Using dMRI tractography to isolate the anterior forceps in the frontal lobe, we demonstrated that underlying microstructure (i.e., fractional anisotropy) of frontal white matter in AIS participants appeared to be altered. This finding is consistent with previous literature investigating frontal white matter in unilateral cerebral palsy reporting lower FA values in anterior cingulate cortex and superior frontal gyrus (Scheck et al., 2015). This is also consistent with the wider ADHD literature that demonstrates lower FA values and higher diffusivity values in frontal white matter of children with ADHD (Konrad & Eickhoff, 2010;Liston et al., 2011;Tremblay et al., 2020;van Ewijk et al., 2012).
Given that these reports use mean metrics extracted from large white matter structures, it is compelling that group differences can still be detected despite using methods with relatively low spatial resolution.

Along-tract analyses and secondary degeneration
Using along-tract analyses, we found spatially specific differences in microstructure metrics in the frontal white matter of children with AIS. Along-tract analyses provided more spatially specific quantification of the spatial extent of possible diaschisis and Wallerian degeneration in areas displaced from the primary lesion allowing for sensitive group comparisons evidenced by significant group by segment interactions in linear mixed models. Our findings suggest that in addition to primary stroke damage to lateral frontal white matter, secondary degeneration of more medial white matter may also underlie group differences in frontal tracts. Such secondary damage of the motor network via diaschisis (Craig et al., 2019;Craig, Olsen et al., 2019;Srivastava et al., 2019) and/or Wallerian degeneration (De Vries et al., 2005;Kirton et al., 2007) has been previously documented in this population. In the current study, some children showed white matter metric values similar to controls while others showed large departures, supporting the idea of heterogeneous and patientspecific secondary degeneration of connected structures in cognitive networks.

Compensatory neuroplasticity
Our findings also highlight areas in the nonlesioned hemisphere that showed higher group mean FA values in both stroke groups compared to controls. This may relate to compensatory developmental neuroplasticity in the nonlesioned hemisphere, though larger studies are needed to more specifically explore associations with cognitive function. Initial studies in this population have demonstrated differences in thalamic (Craig, Carlson et al., 2019) and cerebellar volumes (Craig, Olsen et al., 2019) as well as more complex differences in graph theory metrics quantifying overall structural connectivity of the nonlesioned hemisphere (Craig et al., 2020). Such metrics have been strongly associated with motor function, again suggesting functionally relevant compensatory processes in the nonlesioned hemisphere. We have also documented spatially specific differences in cortical thickness/volumes, gyrification, and sulcal depth (Shinde et al., 2021) as well as myelination (Yu et al., 2018) in the nonlesioned hemisphere after perinatal stroke. Given the heterogeneity of direct stroke-induced damage and the possibility of varying degrees of remote diaschisis and/or neuroplastic compensation in the nonlesioned hemisphere, it is perhaps not surprising that there is large variability in cognitive functioning within this group.
Group differences in frontal white matter were not as apparent in children with PVI, possibly because PVI-induced damage is largely restricted to subcortical periventricular white matter rather than cortex. Because PVI occurs earlier in brain development, there may also be more opportunity for neuroplastic compensation. Indeed, for three regions of the anterior forceps, the PVI group appeared to have higher FA of frontal white matter compared to controls.
Children with PVI are more likely to show motor disabilities (espe-cially of the lower limb) rather than cognitive disabilities given the periventricular location of primary white matter damage and cortical sparing (Dunbar & Kirton, 2019;Kirton et al., 2008). Consistent with this, our sample contained fewer PVI participants (n = 6) than AIS (n = 14) referred for clinical neuropsychological assessment reflecting relatively intact cognition and fewer concerns worthy of referral.

ADHD
Associations between structure and function add additional clinical interest. It is now well established that children with perinatal stroke show a higher prevalence of ADHD compared to children without neurologic injury (Bosenbark et al., 2018;Fuentes et al., 2016;Kirton & deVeber, 2013;Murias et al., 2014;Thomas et al., 2015). We have demonstrated systematic associations between frontal white matter microstructure disruptions (higher diffusivity values) and parent ratings of hyperactivity, inattention, and total ADHD scores consistent with previous studies (Konrad & Eickhoff, 2010;Lawrence et al., 2013;Liston et al., 2011). Lower diffusion anisotropy, leading to higher MD, AD and RD values, is thought to reflect more unrestricted diffusion via underlying differences in axonal membranes, neurofibrils, or cellular density which may in turn be modulated by degree of myelination (Beaulieu, 2002). The apparently disrupted white matter microstructure in frontal anterior forceps as measured by three diffusivity variables appears to be associated with higher parental ADHD ratings.
These findings are consistent with previous literature reporting higher diffusivity values in children with ADHD (Liston et al., 2011;van Ewijk et al., 2012) as well as correlations with parental reports of inattention (Lawrence et al., 2013).
Although MD values showed significant correlations with ADHD ratings, FA did not. It could be that since FA is a complex ratio between diffusion in multiple orientations, it provides a less specific quantification of underlying microstructure compared to other, more directional, microstructure metrics. Further, FA is more affected by underlying white matter structural organization (such as crossing fibers) compared to diffusivity values, which may also contribute to this disparity (van Ewijk et al., 2012). Previous studies investigating ADHD have reported this same pattern where diffusivity metrics show functional correlations and FA does not (Lawrence et al., 2013).

Executive function
Strong correlations were found between age and executive functioning in our perinatal stroke sample despite the use of scaled scores expressed in relation to age-matched normative values. Deficits in cognitive functioning may not become apparent until later in development as parental expectations change as children age (Bosenbark et al., 2018). However, this age correlation could alternatively reflect an altered frontal white matter developmental trajectory for children with perinatal stroke. If the typical trajectory of cognitive development has been arrested, delayed, or prolonged due to early stroke-induced damage (Westmacott et al., 2009) (Scheck et al., 2015), we expected to find that higher FA in frontal white matter would correspond to better function. Further, the two associations we did identify with emotional control and behavioral regulation index were in the strongly positive direction such that higher FA was associated with poorer executive function. As mentioned above, FA is a composite ratio of multiple diffusion orientations and is influenced by many factors such as degree of axonal packing, presence of microtubules, myelination, and crossing fibers (Beaulieu, 2002). The reasons underlying the absence of negative correlations remain unclear; however, they may be related to the sensitivity of FA to multiple underlying tissue properties, its nonspecificity, and the apparent heterogeneity of FA-behavior relationships (Lazari et al., 2020).

Posterior forceps
As hypothesized, the posterior forceps (our comparison tract) appeared to be functionally independent, showing no correlations with executive dysfunction and/or ADHD symptoms. This finding does suggest that differences in frontal white matter after perinatal stroke are functionally specific, at least for the measures presented here, and do not appear to be associated with differences in posterior white matter. Interestingly, posterior white matter did show group differences in volume and underlying microstructure for the AIS group compared to controls. This could be due to direct stroke-induced damage since infarctions of the middle cerebral artery can directly damage posterior visual areas in some cases. In children without direct lesion damage, it also supports the concept of secondary degeneration of areas displaced from primary stroke damage as may be the case with the anterior forceps. These differences also suggest possible deficits in functions mediated by the dorsal visual stream and projections to parietal sensory integration areas after perinatal stroke (Knyazeva, 2013).

Limitations
Certain limitations of the current study must be acknowledged. ROI placement for tract reconstruction was manually performed by one tractographer and thus may have been somewhat subjective compared to automated techniques. This is a ubiquitous challenge for studies using manual tractography but is necessary in this population since the brain anatomy following stroke is typically displaced rendering automated and atlas-based tractography methods ineffective. This limitation was somewhat mitigated by our excellent intrarater reliability of microstructure values. We used tensor-based metrics for white matter microstructure quantification, which are more conservative in streamline selection compared to more complex techniques such as constrained spherical deconvolution (CSD) (Tournier et al., 2004). In addition, using a higher b-value, multishell acquisitions, and better eddy-current distortion corrections may have been more sensitive in detecting differences in WM metrics. Further, more directly modeling free water components of the diffusion signal (Pasternak et al., 2009) would have likely provided additional insight into underlying anatomical architecture. Along-tract segments were in approximately the same position for each participant given that sampling occurred in patient space. Head motion is a challenge for pediatric neuroimaging, especially for participants who may have attentional and cognitive disorders. Those unable to successfully complete an MRI had their scans discarded, ultimately leaving an underrepresentation of more severe cases.
Participants with AIS were overrepresented compared to PVI in the subset of participants with cognitive testing as they were part of a cohort referred for clinical neuropsychological assessment. Comparative measures of ADHD and executive function were not acquired for the control group though population-based normative scales were available for calculating percentiles and T scores. ADHD and executive functioning for the stroke groups were based on parental report, which is informative but may be prone to subjectivity of such parent report.
Prospective, standardized cognitive testing of participants themselves could have been utilized to more directly measure cognitive function and may have provided different associations with white matter metrics. The timing between cognitive assessments and MRI was greater than one year. As neuroimaging metrics can change with experience and age, caution should be taken in interpreting this association when measures are taken at different times.

CONCLUSION
Based on diffusion tensor imaging-based tractography, spatially specific differences in underlying microstructure of the anterior forceps occur in children with perinatal stroke and are associated with stroke type and behavioral outcomes of executive function and attention. Furthering our understanding of associations between cognitive outcomes and the associated underlying brain differences that accompany them may facilitate the advancement of biomarker identification.

ACKNOWLEDGMENT
We thank the children and their families that generously participated in this research.

FUNDING
This study received funding from the Canadian Institutes of Health

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.