Individual differences in children’s global motion sensitivity correlate with TBSS-based measures of the superior longitudinal fasciculus

Abstract

Reduced global motion sensitivity, relative to global static form sensitivity, has been found in children with many neurodevelopmental disorders, leading to the “dorsal stream vulnerability” hypothesis (Braddick et al, Neuropsychologia, 2003). Individual differences in typically developing children’s global motion thresholds have been shown to be associated with variations in specific parietal cortical areas (Braddick et al, J Cog Neuro, 2016). Here, in 125 children aged 5–12 years, we relate individual differences in global motion and form coherence thresholds to fractional anisotropy (FA) in the superior longitudinal fasciculus (SLF), a major fiber tract communicating between parietal lobe and anterior cortical areas. We find a positive correlation between FA of the right SLF and individual children’s sensitivity to global motion coherence, while FA of the left SLF shows a negative correlation. Further analysis of parietal cortical area data shows that this is also asymmetrical, showing a stronger association with global motion sensitivity in the left hemisphere. None of these associations hold for an analogous measure of global form sensitivity. We conclude that a complex pattern of structural asymmetry, including the parietal lobe and the superior longitudinal fasciculus, is specifically linked to the development of sensitivity to global visual motion. This pattern suggests that individual differences in motion sensitivity are primarily linked to parietal brain areas interacting with frontal systems in making decisions on integrated motion signals, rather than in the extra-striate visual areas that perform the initial integration. The basis of motion processing deficits in neurodevelopmental disorders may depend on these same structures.

1. INTRODUCTION

Recent studies of individual differences in measures of visual performance have proved informative for exploring the structure of visual processes (e.g. Goodbourn et al, 2012; Peterzell, 2016). Further insights may be gained by relating these differences in performance to variations in brain structure. The brains of individual children, while following clear common developmental pathways in terms of regional cortical area, thickness, and white matter development, also show marked individual differences in all these measures (e.g. Brown, 2016). Such variations have been correlated with individual differences in behavioral and cognitive function, for example in executive function (Fjell et al, 2012), anxiety (Newman et al, 2015) or intelligence (Fjell et al, 2015). Using this approach with visual perceptuo-cognitive functions, we can consider what brain structures constrain aspects of visual performance in development.

Here we examine structural correlates of children’s performance in the detection of global visual motion, a function dependent on extra-striate processing in the dorsal cortical stream. Global motion sensitivity develops through middle childhood, more slowly than the analogous measure for global static form, with considerable variation between individual children and a dependence on spatio-temporal parameters (for example, Gunn et al, 2002; Atkinson & Braddick, 2005; Hadad, Maurer & Lewis, 2014; Meier & Giaschi, 2014). We have developed child-friendly tests providing comparable measures of global motion and global form, first testing sensitivity for translational motion (‘The Road in the Snowstorm’; Atkinson et al, 1997; Gunn et al, 2002) and in a subsequent version with rotary motion matching form and motion stimuli more closely (‘Find the Ball in the Grass’; Atkinson & Braddick, 2005). Figure 1 illustrates the form and motion components of the ‘Find the Ball in the Grass’ test).

Figure 1. Global form and motion stimuli.

Schematic illustration of the stimuli used in the ‘Find the Ball in the Grass’ tests, shown at 100% coherence. Left: stimulus for motion coherence testing, with the target shown on the right. Right: stimulus for form coherence testing, with the target shown on the left.

Global motion sensitivity is specifically impaired, compared to global form, in a range of genetic and acquired developmental disorders, summarized in table 1, leading us to propose the hypothesis of ‘dorsal stream vulnerability’ (Atkinson et al, 1997; Braddick Atkinson & Wattam-Bell, 2003, Braddick & Atkinson, 2011). These disorders show other deficits of visuo-motor function, spatial cognition, and attention, associated with the functions of the dorsal stream (Atkinson & Braddick, 2011; Kravitz et al, 2011). Understanding the brain correlates of global motion sensitivity, therefore, should help us understand the developmental constraints on these functions and suggest structures that may be associated with its vulnerability in developmental disorders.

Table 1.

Neurodevelopmental disorders showing global motion sensitivity impaired relative to global form sensitivity

Disorder group	Origin	Reference
young Williams Syndrome children	genetic	Atkinson et al (1997, 2003)
adult Williams Syndrome	genetic	Atkinson et al (2006)
autism	possibly genetic	Spencer et at (2002); Simmons et al (2009)
hemiplegic children	acquired	Gunn et al (2002)
developmental dyslexia	?	Hansen et al (2001)
fragile X	genetic	Kogan et al (2004)
very preterm born’	acquired	Atkinson et al, 2007; Taylor et al (2009)
congenital cataract	acquired?	compare Ellemberg et al, (2002) with Lewis et al (2002)
strabismic amblyopia	acquired?	Simmers et al (2005) ; Ho et al (2005)

The opportunity to address these questions comes from an extensive study of a group of children, aged from 5–12 years, who underwent multimodal brain imaging and extensive behavioral assessment, including measurement of global form and motion thresholds, in the PLING study (Pediatric Longitudinal Imaging, Neurocognition & Genetics) (Brown et al, 2012; Fjell et al, 2012; Akshoomoff et al, 2014; Jernigan et al 2016; see http://www.chd.ucsd.edu/research/pling.html) at the University of California, San Diego. We have already reported results from initial analyses on regional measurements of cortical area in this group, which show that good individual performance on a test of global motion sensitivity is associated with relative enlargement of the parietal lobe, especially in the region of the intraparietal sulcus, and relatively smaller area of the occipital lobe (Atkinson et al, 2014; Braddick et al, 2016). Here we extend this analysis to individual differences in fractional anisotropy (FA) in white matter fibre tracts and extend our analysis of regional measurements of cortical areas.

Specifically, we test the hypothesis that individual differences in children’s global motion performance are associated with individual differences in FA in the superior longitudinal fasciculus (SLF), a major fibre bundle with extensive connections in the parietal lobe and providing two-way communication with frontal and prefrontal areas, including supplementary motor and premotor areas (Kamali et al, 2014). Given the relationship of global motion sensitivity to parietal area, we expect that this tract may play an important role in transmitting motion information to anterior cortical areas, and perhaps in top-down modulation of motion processing in parietal and extra-striate areas. It is possible that such modulation, a form of attentional control, may play a significant role in individual abilities to detect global motion. There is long-standing evidence that spatial attention has a lateralized brain basis in the right hemisphere (e.g. Mesulam, 1981) and more specifically, parieto-frontal attentional networks, including the superior longitudinal fasciculus, are known to be lateralized (e.g. Szczepanski et al, 2010; Thiebaut de Schotten et al, 2011). We therefore examined relations with the right and left superior longitudinal fasciculiincluded separately in our statistical models.

2. METHODS

2.1 Participants

The participants were drawn from a sample of children making their first visits for data collection in the PLING study at the University of California, San Diego (UCSD). In addition to measurements of global form and motion sensitivity, participants completed an extensive battery of cognitive tests, as well as a structural magnetic resonance imaging (MRI) session in which diffusion weighted images were acquired. The work was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and the human research protections program and institutional review board at UCSD approved all experimental and consenting procedures. Written parental informed consent was obtained for all participants and child assent from those 7 years and older. Participants were screened for history of major developmental, psychiatric, or neurological disorders, brain injury, or other medical conditions that affect development. Only participants aged between 5:0 and 12:11 years, who completed two measurements of both global motion and global form sensitivity, with neuroimaging data including fractional anisotropy images that passed a quality check, were included.

Data are presented from 125 children whose diffusion weighted imaging data met the quality control criteria, a subset of the 154 whose cortical area measurements were analyzed in Braddick et al (2016). They included 66 males and 59 females, mean age 7.53 yr (s.d.= 1.76 yr); numbers in each 1-year band were: 5 yr+ N=31; 6 yr+ N=23; 7 yr+ N=18; 8 yr+ N=27; 9 yr+ N=13; 10 yr+ N=13.

2.2 Global form and motion testing: the ‘Find the ball in the grass’ test

Global motion and form thresholds were determined as the minimum percentage of coherently organized elements, embedded among random noise elements, for which children could detect global structure. The test was similar to that of Gunn et al (2002) but with concentric stimulus displays (Atkinson & Braddick, 2005) on a laptop computer screen, designed to make the form and motion tasks as comparable as possible. Participants had to report whether a circular region, ‘the Ball’, containing concentrically organized short arcs or trajectories of moving dots, appeared on the left or right of center in a background of randomly oriented elements.

The display subtended 25 × 18 deg arc at the viewing distance of 50 cm. The global motion display contained 3000 dots each 11 min arc diameter, moving at 4.1 deg/sec. Each dot had a lifetime of 8 frames (133 msec) after which it disappeared from the screen. Within the circular target region (the “Ball”), diameter 9.5 deg arc centered 6.3 deg left or right of screen center, coherent dots moved in concentric circular paths. The percentage of dots sharing this coherent motion varied from trial to trial as described below. The remaining dots within this region, and all the dots elsewhere on the screen, moved in randomly oriented arcs whose curvature followed the same distribution as the population of coherently moving dot trajectories.

The global form display contained 3000 stationary arc segments, each 8 min arc wide × 42 min arc long. Within the same circular “Ball” regions as in the global motion display, the coherent arcs were oriented to be concentric. The percentage of arcs sharing this coherent alignment varied according to the same adaptive procedure as for motion. The remaining arcs within this region, and all the arcs elsewhere on the screen, were randomly oriented arcs with the same distribution of curvature as the coherently oriented arcs. Both displays are illustrated in Figure 1.

Participants completed four test runs, alternating runs with the global form and the global motion display, with the starting test randomized across participants. On each trial, the structured “Ball” region was presented randomly on the left or right of centre, and the child was asked to point to the side that contained the “Ball” (circular pattern), or, for older children, to press the corresponding arrow key on the keyboard. The display terminated if the child had not responded within 10 seconds, although most responses were made in the first few seconds. The child’s response terminated the trial. Each run began with coherence fixed at 100%, and these trials were continued until the tester was satisfied that the child understood the task. In the following test phase, the coherence level of the target region was varied according to the Psi adaptive procedure (Kontsevich & Tyler, 1999) which uses a Bayesian approach to place each trial at the point where it will give the most information about the 2-dimensional posterior probability distribution of the threshold and slope of the psychometric function. The estimated threshold was the mean threshold from this distribution, after the completion of 30 trials. Since the adaptive procedure leads to difficult decisions as stimuli are delivered increasingly close to the individual’s threshold, the children’s motivation was maintained by delivering an easily visible stimulus (100% coherence) on every sixth trial. In over 95% of occasions, the responses to these ‘catch’ trials was correct; when it was not correct the child was reminded to indicate the side containing the circular pattern. No children were eliminated for repeated errors on these ‘catch’ trials.

The thresholds used in analysis were the mean of two values from a child’s two test runs with the specific stimulus (global motion or global form)

2.3 Neuroimaging and analysis: MRI scanning protocol

The neuroimaging protocol was as described in Jernigan et al (2016a). A standardized multiple-modality high-resolution structural MRI protocol was implemented, involving 3D T1-weighted volumes and a set of diffusion-weighted scans, on a GE 3T Signa HD× scanner and a 3T Discovery 750× scanner (GE Healthcare) using eight-channel phased array head coils. The protocol included a conventional three-plane localizer, a sagittal 3D inversion recovery spoiled gradient echo T1-weighted volume optimized for maximum gray/white matter contrast (echo time = 3.5 ms, repetition time = 8.1 ms, inversion time = 640 ms, flip angle = 8°, receiver bandwidth = ±31.25 kHz, FOV = 24 cm, frequency =256, phase = 192, slice thickness = 1.2 mm), and two axial 2D diffusion tensor imaging (DTI) pepolar scans (30-directions b-value= 1,000, TE = 83 ms, TR = 13,600 ms, frequency = 96,phase = 96, slice thickness = 2.5 mm, FOV =24 cm).

2.4 Analysis of fractional anisotropy of white matter

Our methods for defining fractional anisotropy (FA) of the Superior Longitudinal Fasciculus followed closely those of Vestergaard et al (2011). All fibre tracts were spatially normalized and aligned across all subjects using Tract-Based Spatial Statistics (Smith et al, 2006), a module in FSL 5.0.2 (Smith et al, 2004). The FA images of all subjects were aligned into a common space by implementing the non-linear registration tool FNIRT (Andersson et al, 2007a,b). Next, a subset group of 50 participants were selected to be representative of the whole group matched in distribution of age and sex. For these 50 participants, all FA images were non-linearly registered to the FA images of every other subject within this group, in order to find the group’s most representative FA image. Using affine registration, the most representative FA image was then aligned to MNI space and subsequently the subset of 50 participants were aligned to this image and transformed into 1 mm³ MNI space. The mean FA image for the subset group of 50 participants was created, and formed the study-specific target to which the entire dataset was aligned, and transformed into 1 mm³ MNI space. Next, a cross-subject mean FA image from the entire dataset was created and thinned to generate the mean FA skeleton for the entire dataset, representing the centers of all tracts common to the group. The mean FA skeleton for the entire dataset was thresholded at FA > 0.2 and contained approximately 119,017 1 mm³ interpolated voxels. All participants’ aligned FA data were then projected onto this skeleton, by locating the voxels with the highest local FA value in the direction perpendicular to the tracts in the skeleton and assigning this value to the subject’s skeleton at the given standardized location. Such projection yields a mapping of each voxel location in the skeleton to a specific voxel in the individual participants’ FA maps. This process accounts for residual misalignments between individuals after the initial registration, and minimizes any systematic between-participants differences in tract location. An eigenvalue color-coded map corresponding to the target FA map was created by applying the affine transformation generated when aligning the most representative FA image to MNI space to corresponding primary eigenvector (v₁) image using the FSL “vecreg” command.

The Superior Longitudinal Fasciculus region-of-interest (ROI) was delimited for both the left and right hemisphere on the mean FA skeleton overlaid on the target FA map. It was defined using the color-coded target FA map in accordance with the MRI atlas from Mori et al. (2005). The SLF ROI contained the “stalk” or core of the tract, which extended from superior parietal cortex to dorsolateral prefrontal cortex, and limited to voxels with anterior-posterior oriented primary eigenvectors. Of the distinct branches of the SLF (Makris et al, 2005), it corresponds most closely to SLF II, although it tends to exclude angled or bending fibres because of the restriction to anterior-posterior eigenvectors. Figure 2 illustrates the region of interest so defined. The left SLF contained 577 voxels and the right SLF contained 522 voxels. The mean FA within each region of interest was extracted for both the left and the right SLF as well as whole skeleton FA, an estimate of global white matter FA, for all participants. The mean SLF in the bilateral SLF was found to be normally distributed across participants (W = 0.99, p-value = 0.84 on the Shapiro-Wilk normality test)

Figure 2.

Depiction of region of interest defined for the superior longitudinal fasciculus (SLF). [A] depicts an axial slice at MNI co-ordinate z=36; [B] depicts a sagittal slice of the right hemisphere at x=37. The images show the mean FA skeleton in yellow, overlaid on the color-coded DTI maps where green indicates diffusion in the anterior-posterior direction, blue in the inferior-superior direction, and red in a lateral direction. The red region superimposed on the skeleton is the defined region of interest (anterior-posterior diffusion was one of the criteria used- see text).

2.4 Area measures of individual cortical lobes

Results reported here for relation to cortical area were obtained from the procedures described in Braddick et al (2016). In particular the areas of the occipital and parietal lobes were defined from a genetically informed, cortical parcellation scheme, which used a fuzzy clustering method to analyze the matrix of genetic correlations among vertex-wise estimates of cortical surface expansion in a sample of monozygotic and dizygotic twins (Chen et al, 2012).

2.5 Statistical analysis

The relationships between the visual performance measures of global form and motion, and the anatomical measures of fractional anisotropy and of lobe area, were analyzed by multiple regression models fitted by a least-squares criterion using the software package JMP Pro 12 (SAS Institute, Cary NC). Children’s global motion thresholds or global form thresholds were used as the dependent variable, with age, age² gender, and MRI scanner as predictors, as well as the anatomical measures of interest. These analyses yielded tables of parameter estimates with t-values and significance levels for each effect: examples of these are presented in tables 2–5.

Table 2.

Regression models for motion and form coherence thresholds against fractional anisotropy (FA) in left and right superior longitudinal fasciculus (SLF). Note that a negative t-value signifies a positive association with high sensitivity to global motion or form.

	Motion coherence threshold		Form coherence threshold
Term	t ratio	p > |t|	t ratio	p > |t|
Intercept	4.98	<0.0001****	2.07	0.040*
Age	−7.75	<0.0001****	−4.34	<0.0001****
Age* age	3.57	<0.0005***	1.99	0.049*
Gender	0.34	0.733	0.30	0.765
Left SLF FA	2.43	0.017*	0.45	0.652
Right SLF FA	− 2.99	0.003**	0.10	0.917

^****= p ≤ 0.0001

^***= p ≤ 0.001

^**= p ≤ 0.01

^*= p ≤ 0.05

Table 5.

Regression models for motion coherence thresholds against FA of left and right SLF, and bilateral parietal area.

Term	t ratio	Prob > |t|
Intercept	5.7	<0.0001****
Age	−7.78	<0.0001****
Age*age	4.00	0.0001****
Gender	−1.14	0.258
Bilateral parietal area	−2.57	0.011*
Right SLF FA	−2.76	0.0068**
Left SLF FA	2.49	0.014*

To explore the spatial distribution of relationships beyond the hypothesised link between SLF and global motion, separate effect size maps were also constructed, to showing the t-value of the correlations between global motion thresholds and FA, and global form thresholds and FA, adjusted for age, age², and gender. The t-maps were generated in FSL using whole brain linear regression across the whole skeleton with the global motion measure, age, age², and gender as predictors and FA as the dependent variable. Voxel level t-value ranges for both positive and negative correlations (df = 120), uncorrected for multiple comparisons, are indicated by a color code described in the legend for Figure 6.

Figure 6.

Effect-size maps of the associations between global thresholds and fractional ansiotropy (FA), adjusted for age, age², and gender. The effect-size maps are overlaid on the target FA image. The center panel shows the mapping of colors to values of the t-statistics. All voxels on the skeleton are colored. |t| > 1.980 corresponds to an uncorrected p value of < 0.05, and |t| > 2.617 to p < 0.01. The MNI coordinates for each section are indicated on the image, along with R indicating the right hemisphere and A, P the anterior-posterior orientation of the axial section [F]. Red arrows indicate clusters of voxels discussed in section 3.3. Negative associations mean that high FA was associated with high sensitivity (low threshold).

[A]–[D] show effect sizes for the correlation of global motion thresholds with FA: [A], [B]: cluster of voxels in the right superior longitudinal fasciculus showing a negative association with motion coherence threshold; [C], [D]: clusters of voxels in right anterior and left superior corona radiata, showing a negative association with motion coherence thresholds

[E]–[G] show effect sizes for the correlation of global form thresholds with FA: [E], [F]: cluster of voxels in right anterior corona radiata showing a negative association with form coherence thresholds; [G] cluster of voxels in the cerebellum showing a positive association with form coherence thresholds.

RESULTS

3.1 Motion and form coherence thresholds as a function of age

As expected, there was a marked improvement over the age range of this sample in sensitivity both to global form and to global motion. Data from the present sample are illustrated in Figure 3. In line with earlier findings (Gunn et al 2002, Atkinson & Braddick 2005, Braddick et al, 2016), these measurements showed a steeper improvement with age for global motion than for form, especially in the younger part of the range, and a greater individual variability for motion than for form. The approach to an asymptotic value around 8–9 years is consistent with these and other results (e.g. Braunitzer et al, 2012; see also review by Hadad et al, 2015). However, despite the different age functions, global form and motion thresholds had a substantial shared variance: when the effect of age was removed by analysing the studentized residuals for each threshold in a regression against age and gender, individual form and motion thresholds showed a correlation of r = 0.597.

Figure 3.

Individual children’s coherence thresholds for global motion (red) and global form (blue), plotted as a function of age. Each point is the mean of two measurements. The lines are quadratic fits to the data points. N = 125

3.2 Relation of coherence thresholds to fractional anisotropy in the superior longitudinal fasciculus

Table 2 summarises statistical models for the prediction of motion and form coherence thresholds from the fractional anisotropy measures of the superior longitudinal fasciculus. The right and left SLF were opposite predictors of global motion sensitivity: higher fractional anisotropy in the right-hemisphere’s superior longitudinal fasciculus is associated with lower motion thresholds (high sensitivity), while higher fractional anisotropy in the left hemisphere’s superior longitudinal fasciculus was associated with higher motion thresholds (low sensitivity). Figure 4 shows leverage plots, (i.e. the unique effects of left and right fractional anisotropy respectively on global motion thresholds). In contrast to these results for global motion, no such relationship was found with global form sensitivity. Associations with global motion sensitivity but not global form were similarly found in the structural measures of cortical area reported by Braddick et al (2016).

Figure 4.

Leverage plots of global motion thresholds against fractional anisotropy of the left and right superior longitudinal fasciculus. Leverage plots indicate the relationship between residuals when other variables in the model (age, age², and gender) are allowed for.

Similar models were tested for prediction by the whole skeleton fractional anisotropy. These yielded no significant associations with either global motion or global form sensitivity, either when entered alone into the model or alongside fractional anisotropy of right and left superior longitudinal fasciculus (Table 3). The latter result indicates that the associations with the left and right superior longitudinal fasciculus were anatomically specific and not a reflection of overall white matter maturation.

Table 3.

Regression model for motion coherence thresholds against fractional anisotropy (FA) in left and right superior longitudinal fasciculus (SLF) and in the overall white matter skeleton. Note that a negative t-value signifies a positive association with high sensitivity to global motion or form.

	Motion coherence threshold
Term	t ratio	p > |t|
Intercept	3.14	<0.0021**
Age	−6.72	<0.0001***
Age* age	3.56	<0.0005***
Gender	0.15	0.881
Left SLF FA	2.54	0.012*
Right SLF FA	−2.59	0.011**
Total skeleton FA	−0.83	0.406

^***= p ≤ 0.001

^**= p ≤ 0.01

^*= p ≤ 0.05

Since global motion performance was strongly associated with age in the range studied, one might ask whether the relationship with fractional anisotropy in the left and right superior longitudinal fasciculus reflects differential maturation between individual children, that is, was the asymmetry between the tracts a feature of the more mature brain? This appears unlikely. The average fractional anisotropy of the left and right increased with age (t= 4.14, p < 0.0001 in a model including age, age², and gender: see Figure 5a). However an analogous model for the ratio (left SLF FA):(right SLF FA) showed no age effect and a very flat relation (Figure 5b). Over the group as a whole, there was no significant difference between fractional anisotropy values for the left and right superior longitudinal fasciculi (means 0.497 and 0.499 respectively; matched-pairs t=1.206, p = 0.23)

Figure 5.

(a) scatter plot of mean fractional anisotropy of left and right superior longitudinal fasciculus for individual children, as a function of age. Dotted line is a quadratic fit (regression highly significant – see text); (b) scatter plot of the ratio of fractional anisotropy of left vs right superior longitudinal fasciculus for individual children, as a function of age. Dotted line is a quadratic fit (no significant age variation)

3.3 Associations with coherence thresholds in whole-brain fractional anisotropy data

The primary goal of this study was to test the specific anatomical hypotheses about the relationship between microstructure of white matter in the superior longitudinal fasciculus, and variations in children’s sensitivity to global motion. Therefore neither a whole-brain, voxel-wise analysis of the effects (appropriately adjusted for multiple comparisons), nor a restricted voxel-wise analysis with small volume correction was considered appropriate for testing this a priori hypothesis. However, since the analyses we used produced estimates of the effect size at each voxel, we have provided visualization of the effect-size maps showing positive and negative associations between FA and motion coherence threshold, adjusted for age, age2, and gender (Figure 6A–D). A similar effect size map for the positive and negative associations between FA and form coherence threshold is presented in Figure 6E–G.

As well as the right superior longitudinal fasciculus, which was tested in our original hypothesis, clusters of voxels with negative association with motion coherence threshold were found in the corona radiata (Figures 6C–D). The region in the right anterior corona radiata also overlaps strongly with voxels showing a negative association with form coherence thresholds (Figure 6E, F). Finally, positive association with form coherence thresholds (i.e high fractional anisotropy linked to low global form sensitivity) occurred in a midline region of the cerebellum (Figure 6G). However, in no case did a voxel meet the criterion for significance at p<0.05 when corrected for multiple comparisons. For the motion threshold, 1,403 voxels had a t-value above 1.980 and 5,183 voxels had a t-value below −1.980 (uncorrected p<0.05), comprising 5.53% of the voxels in the skeleton. For the form threshold, 1,811 voxels had a t-value above 1.980 and 3,266 voxels had a t-value below −1.980 (uncorrected p<0.05), comprising 4.27% of the voxels in the skeleton.

3.4 Superior longitudinal fasciculus and cortical area

The asymmetry in the relation between motion thresholds and left and right superior longitudinal fasciculus prompted us to re-examine the association of these thresholds with the area of cortical lobes. The analysis in Braddick et al (2016) examined relationships with the total area of each lobe, combining left and right hemispheres. We have now tested regression models of these data in which left and right parietal and occipital lobes are examined separately. The results (Tables 4a and 4b) showed that the effects of area (lower motion thresholds with high parietal area, higher motion thresholds with high occipital area) were in the same direction for left and right hemispheres. However, the effects of parietal and occipital area previously reported in the bilateral data were driven most strongly by the areas of these lobes in the left hemisphere. Thus the overall results reflect a complex pattern of structural asymmetry, in which cortical area effects were dominated by the left hemisphere, where SLF fractional anisotropy shows a negative relation with motion performance, and the positive association of the right superior longitudinal fasciculus was accompanied by a weaker relationship of lobe areas to motion performance in that hemisphere.

Table 4a.

Regression models for motion coherence thresholds against area of the left occipital and left parietal cortical lobes, with total cortical area of the left hemisphere included in the model. Note that a negative t-value signifies a positive association with high sensitivity to global motion or form.

Term	t ratio	p > |t|		t ratio	p > |t|
Intercept	7.04	<0.0001****		7.46	<0.0001****
Age	−7.30	<0.0001****		−8.06	<0.0001****
Age* age	3.81	0.0002***		3.95	0.0001****
Gender	−0.92	0.359		−0.87	0.384
L occipital area	2.45	0.016*	L parietal area	−2.14	0.0348*
L total cortical area	−3.18	0.0019**	L total cortical area	1.24	0.219

^****= p ≤ 0.0001

^***= p ≤ 0.001

^**= p ≤ 0.01

^*= p ≤ 0.05

Table 4b.

As table 4a, but regression models for motion coherence thresholds against cortical areas in the right hemisphere.

term	t ratio	p > |t|		t ratio	p > |t|
Intercept	7.12	<0.0001****		7.38	<0.0001****
Age	−7.61	<0.0001****		−7.67	<0.0001****
Age* age	3.37	0.001***		3.87	0.002**
Gender	−0.86	0.390		−1.01	0.315
R occipital area	2.15	0.034*	R parietal area	−1.07	0.286
R total cortical area	−2.85	0.005**	R total cortical area	0.38	0 .707

^****= p ≤ 0.0001

^***= p ≤ 0.001

^**= p ≤ 0.01

^*= p ≤ 0.05

Do these effects, nonetheless, reflect a common phenotype of asymmetrical brain development? This can be tested by a regression model which includes bilateral parietal area, together with fractional anisotropy in the right and the left superior longitudinal fasciculus (Table 5). Each of these structural measures made a significant, independent contribution to the prediction of global motion thresholds. These relationships, therefore, cannot depend primarily of shared variance between the structural measures. In fact, although fractional anisotropy of the left and right superior longitudinal fasciculi showed substantial shared variance, as did and left and right areas of parietal and occipital lobes, the correlations between fractional anisotropy of the superior longitudinal fasciculus and cortical area measures were very weak when age, overall area and whole skeleton fractional anisotropy are taken into account (Table 6).

Table 6.

Matrix of intercorrelations between structural measures. All measures residualised for age and gender. Area measures are also residualised for total cortical area; FA measures are residualised for whole skeleton FA. Significance levels, where not stated, are all p > 0.05

	R parietal area	L occipital area	R occipital area	L SLF FA	R SLF FA
L parietal area	0.356 P < 0.001	−0.408 p < 0.001	−0.400 p < 0.001	−0.111	−0.031
R parietal area		−0.240 p<0.01	−0.268 p < 0.01	−0.173 p = 0.532	−0.045
L occipital area			0.783 p < 0.001	0.173 p = 0.539	0.097
R occipital area				0.090	−0.034
L SLF FA					0.698 p < 0.001

3. DISCUSSION

Our earlier analysis (Braddick et al, 2016) showed that children’s global motion performance was related to variations in cortical structure in the parietal lobe, particularly in the region of the intraparietal sulcus (IPS). On this basis, we hypothesised that the superior longitudinal fasciculus, through which this area interacts with anterior cerebral structures, might show a similar relation to visual motion processing. Our results confirmed that such a link exists, but that it is complex, with opposite relationships between the structure of the superior longitudinal fasciculus and motion sensitivity in the two hemispheres. Furthermore, on closer examination, the relation with parietal and occipital cortical areas also was found to be asymmetrical. The sign was the same in both hemispheres, but was markedly stronger for the left hemisphere. Thus the brain of the child who has high sensitivity to global motion shows a complex pattern of asymmetry relative to the brain of a child with lower performance, although the different components of this asymmetry contribute independently, at least to some degree.

Csete et al (2014) have also reported a relationship between motion coherence threshold and fractional anisotropy for some voxels in a number of white matter regions, in a smaller group of adult participants. They did not test specific tracts, although they state that some of the locations they identify lie within the medial branch of the left superior longitudinal fasciculus. For all the voxels they reported, the correlation of fractional anisotropy with motion thresholds is positive – that is, higher fractional anisotropy goes with reduced sensitivity, in line with our own finding for the left SLF.

We tested a specific hypothesis of association between global visual performance and the superior longitudinal fasciculus. The whole-skeleton analysis of effect size allows us to examine how this association may be distributed in white matter. Additional locations for association were found in the corona radiata. These included a region of the right anterior corona radiata which was associated with both global motion and global form sensitivity, and so may play a different role to the apparently motion-specific structure in the superior longitudinal fasciculus. While the corona radiata is most often thought of as the fan of fibres which converge into the internal capsule and project to subcortical and spinal targets, it does include cortico-cortical tracts also (Wakana et al, 2004). Fractional anisotropy in the corona radiata has been associated with individual variations in a number of cognitive functions, including children’s numerical abilities (Matejko & Ansari, 2015), cognitive control (Seghete et al, 2013; Chaddock-Heyman et al (2013); literacy (Myers et al, 2014), and IQ scores (Chiang et al, 2009). This structure may therefore be associated rather broadly with the development of information processing abilities, and its specific role in brain networks subserving these abilities, and in global visual processes specifically, is unclear. In any case, given that this and other relationships shown in the effect size maps have only been shown with uncorrected significance levels, they must be seen as pointers for possible further investigation rather than established findings.

4.1 Asymmetries in motion processing

A number of reports find that individual participants show hemispheric asymmetries in cerebral responses to motion, for example, Patzwahl et al, (1994), Hollants-Gilhuijs et al (2000), and Nakamura et al, (2003), which they ascribe to area V5/MT. However, there does not appear to be any consistent difference across individuals between left and right hemispheres in these responses, which might be associated with the asymmetric anatomical relationships we find with motion coherence thresholds. An exception is an overall right-hemisphere advantage reported by Patzwahl et al (1994), which is ascribed by them to “[the special role of the right hemisphere] attentional system amplifying the activation of the right parietal cortex”. In contrast, Hollants-Gilhuijs et al (1998) found a behavioral advantage for right-field motion detection (implying left-hemisphere advantage) in 6- to 16-year-old children (but not in adults), which they ascribed to differential maturation of extra-striate areas in the two hemispheres. A number of other visuo-spatial tasks have been found to show hemispheric asymmetries, not necessarily in this direction (e.g. Atkinson & Egeth, 1973). Since our task involved intermixed left- and right-field stimuli, and the children could freely fixate while making their decision, our data provide no evidence on half-field differences and do not allow us to examine whether such a functional asymmetry is linked to the structural correlates we find. This is a possible question for future studies using a task which requires central fixation and brief stimulus presentations in either left or right field, which would not however be practical with children across the age range of the present study.

4.2 Attentional and cognitive role of the superior longitudinal fasciculus

Our findings present a specific link of structure of the superior longitudinal fasciculus to the processing of motion information. However, neuroimaging and lesion studies in adults have identified a number of broader aspects of cognitive function that are associated with this tract, especially within the domain of spatial cognition and attention. Bennett et al (2012), Mayer and Vuong (2014) and Chechlacz et al (2015) report correlations of fractional anisotropy in the superior longitudinal fasciculus with various measures of visuo-spatial attention; Rodriguez-Herreros et al (2015) find that fractional anisotropy in SLF-2 correlates with the individual resistance to TMS disruption in visually guided reaching. Damage to the superior longitudinal fasciculus is also a key cause of the spatial neglect syndrome (e.g. Shinoura et al, 2009; Ptak & Schnider, 2010; Chechlacz et al, 2012; Lunven & Bartolomeo, 2016).

Other associations have been found between fractional anisotropy of the superior longitudinal fasciculus, and working memory (Vestergaard et al, 2011; Rizio & Diaz, 2016), executive function and sustained attention in children (Klaborg et al, 2013; Urger et al, 2015), and children’s reading development (Travis et al, 2016; Wang et al, 2016).

Fractional anisotropy in the superior longitudinal fasciculus is also reduced compared to controls in several neurodevelopmental disorders, including cerebral palsy, where it is correlated with IQ (Ballester-Plané et al, 2016) and amblyopia, where it is correlated with visual acuity (Li et al, 2015). It is also correlated with response time variability in attention-deficit hyperactivity disorder (Wolfers et al, 2015). However in Williams Syndrome, fractional anisotropy of the superior longitudinal fasciculus is increased relative to controls, and high fractional anisotropy is associated with poor visuospatial performance (Hoeft et al, 2007).

More broadly, early disorders of cerebral white matter are a major feature of neonatal brain injury associated with deficits of early attention and dorsal stream function in preterm born children (Atkinson et al, 2008) and of early brain development in Williams syndrome (Mercuri et al, 1997).

The networks associated with spatial attention are not symmetrical: Thiebaut de Schotten et al (2011) describe a “lateralized brain network for visuospatial attention” in which the volume of right SLF II is associated with left bias in line bisection and with faster left field detection. These results, and the wider findings of attention-related functions suggest that our finding of fractional anisotropy of the right superior longitudinal fasciculus associated with good motion performance may be linked to deployment of attention in the motion processing task. Studies of functional connectivity (Friston & Büchel, 2000) have shown that activity in the parietal lobe is associated with attentional modulation of the transmission of motion information from V1/V2 to V5/MT. However, it should be noted that the global form task, which shares many of the attention demands of the global motion task, shows none of the same relations with brain structure.

The relationship between fractional anisotropy and function is complex, since a number of different tissue properties will contribute to the measured FA (Beaulieu, 2002). It is often assumed that fractional anisotropy reflects the integrity of tract organization, in terms of the parallelism and packing of axons which restricts radial diffusion and hence increases fractional anisotropy; myelination is another functionally important aspect of brain development which similarly increases fractional anisotropy. However, some factors which reduce fractional anisotropy may be associated with enhanced function: for example increased axonal diameter which may be responsible for the association seen between faster choice reaction time and reduced fractional anisotropy in the optic radiation (Tuch et al, 2005). Hoeft et al (2007), discussing the increased fractional anisotropy and its association with poor visuospatial test performance in Williams Syndrome, suggest that this higher fractional anisotropy may be associated with reduced branching in the tract, a proposed characteristic of this neurodevelopmental disorder (Eckert et al, 2006). Thus we cannot necessarily assume that a positive correlation with high fractional anisotropy means that a particular fibre tract contributes positively to motion processing in development (or vice versa for a negative correlation). Furthermore, the direction of causality between behavioral performance and structural differences remains uncertain.

4. CONCLUSIONS

The finding that the structure of the superior longitudinal fasciculus is related to individual variations in children’s global motion processing adds support to two conclusions from our earlier findings on their relation to regional cortical area (Braddick et al, 2016).

First, these individual differences do not necessarily reflect the stage, in extrastriate areas such as MT/V5 or V3a, where local motion signals are initially integrated to provide information about globally coherent motion (Mikami, Newsome & Wurtz, 1986; Newsome & Paré, 1988; Braddick et al, 2001), although we cannot exclude localized variations in these areas. Instead, the correlations with individual differences in global motion sensitivity are provided by variations in the area of the parietal lobe (which receives input from extrastriate motion areas), and in the superior longitudinal fasciculus through which parietal cortex communicates with anterior cerebral areas. These results suggest that the critical point for determining these individual differences may be the levels where sensory evidence for global motion is accumulated for perceptual decision-making, and at which these decisions are communicated to response systems (Shadlen & Newsome, 2001; Hanks et al, 2006). The role of attention systems in contributing to this performance has been discussed above; top-down modulation of parietal function may play an important role in individual differences.

Second, we identify structural correlates of global motion sensitivity but have not yet confirmed any for sensitivity to static global form. Impaired motion rather than form sensitivity provides the widespread signature of neurodevelopmental disorders (“dorsal stream vulnerability” – Braddick Atkinson & Wattam-Bell, 2003). It appears that even within the range of typical development, it is again motion processing which provides the more sensitive index of variations in brain development.

The developmental course of these structure-function relationships is still unknown. Future work using longitudinal data sets from this and other groups of children may help to show whether early structural differences predict later functional development. Studies of genetic associations with motion thresholds and with brain structure (e.g. Schork, 2012) may also help to understand the causal pathways that determine these patterns of individual differences.

Highlights.

• Children’s individual motion sensitivity correlates with fractional anisotropy of the Superior Longitudinal Fasciculus (SLF)

• This correlation is opposite in left and right hemispheres

• Individual differences in motion processing depend on levels beyond V5/MT

• Differences in SLF and parietal structures may contribute to “dorsal stream vulnerability”

Abbreviations

TBSS

tract-based spatial statistics

FA

fractional anisotropy

SLF

superior longitudinal fasciculus

FNIRT

FMRIB nonlinear image registration tool

MNI

Montreal Neurological Institute

IPS

intraparietal sulcus

TMS

transcranial magnetic stimulation