Dr Richard E Frye at Department of Pediatrics, University of Texas Health Science Center, 7000 Fannin, UCT 2478, Houston, TX 77030, USA. E-mail: firstname.lastname@example.org
Aim To understand the relationship between cognition and white-matter structure in adolescents born preterm without obvious brain injury.
Methods Thirty-two adolescents from a longitudinal study of child development were selected according to risk of developmental disorders at birth (born at term: eight males, five females; median age 16y 1mo, interquartile range 10mo; low risk preterm: four males, five females, median age 16y, range 4mo; high risk preterm: three males, seven females, median age 16y 2mo, range 1y 2mo) and reading ability (good: three males, eight females, median age 16y, range 7mo; average: six males, three females, median age 16y 10mo, range 1y; poor: six males, six females, median age 16y, range 6mo). Preterm birth was defined as a gestational age of 36 weeks or less and a birthweight of 1600g or less. All participants had normal clinical neuroimaging findings. We examined fractional anisotropy, radial diffusivity, and volume of three major white-matter fasciculi. The relationship between structural measures and birth risk, hemisphere, and cognitive ability (attention, lexical and sublexical decoding, auditory phonological awareness, and processing speed) were analysed using mixed-model regression.
Results Left-hemisphere superior longitudinal fasciculus (SLF) fractional anisotropy and radial diffusivity were linked to reading-related skills (fractional anisotropy vs letter–word identification, r(30)=−0.37, p<0.05; fractional anisotropy vs phoneme reversal, r(30)=−0.34, p=0.05; radial diffusivity vs letter–word identification, r(30)=0.31, p<0.10; radial diffusivity vs phoneme reversal, r(30)=0.40, p<0.05), whereas right-hemisphere SLF fractional anisotropy was related to attention skills (fractional anisotropy vs inattentiveness, r(30)=−0.38, p<0.05). SLF volume decreased as these skills declined for adolescents born preterm (volume vs phoneme reversal, r(17)=0.58, p<0.01; volume vs inattentiveness, r(17)=−0.69, p<0.01), but not for those born at term.
Interpretation The relationship between cognitive skills and SLF volume suggests that in adolescents born preterm, cryptic white-matter injury may exist, possibly related to oligodendrocyte or axonal loss, despite normal clinical neuroimaging.
Children born preterm are more likely than those born at term to require special assistance and to be diagnosed with a specific learning disability despite adequate intelligence. For example, school-age children who were born preterm have deficits in executive function,1 visuospatial skills,1 and early language development.2 The relationship between brain anatomy and neuropsychological and achievement deficits is unclear.
Over the past two decades we have studied the development of a large cohort of children who were born at term and preterm. Children born preterm were recruited to represent those who were at low risk and high risk of developmental difficulties based on their medical complications during the neonatal period. We recently investigated the relationship between verbal, nonverbal, and executive-function skills and reading during the 3rd, 5th, and 7th school grades (i.e. children aged around 8y, 10–11y, and 12–13y) in this cohort.3 Participants were divided into three groups of reading ability based on the levels and growth in the performance on the Woodcock–Johnson Test of Achievement Word Attack subtest during the study period. Children born preterm were found to be no more likely to be poor readers, but children born preterm at high risk with poor reading ability were found to perform poorly on executive-function tasks, particularly on the inattention index of a continuous performance task.4 This suggests that poor readers born preterm at high risk may represent a group of children with more severe cognitive dysfunction.
Anatomic neuroimaging has revealed selective disturbances in the microstructure of the deep white-matter tracts and the longitudinal fasciculi in children born preterm who do not show obvious brain injury on prenatal, perinatal, or neonatal ultrasound examination.5 Attention deficit in preterm children has been linked to the microstructure of the inferior and superior longitudinal fasciculi, and inattention has been correlated with the microstructure of the superior fasciculi.5 The superior longitudinal fasciculus (SLF) may be especially important during child development, because in addition to being related to inattention5 it connects cortical regions of the brain involved in reading,6 and diffusion tensor imaging has implicated white-matter disorganization of this fasciculus in individuals with reading disability.7 Thus, we believe that this pathway may be specifically affected in poor readers born preterm at high risk of developmental disorders. Such a finding would support our behavioral link between poor reading and executive dysfunction in children born preterm at high risk. In this article we use an atlas-based diffusion tensor imaging algorithm to examine the relationship between the microstructure and macrostructure of the longitudinal fasciculi and birth risk (i.e. term, preterm low risk, and preterm high risk) and two important aspects of cognitive function, reading-related skills and attention, in a subsample of our previously studied cohort.3
The original cohort
Our original longitudinal cohort contained 360 children recruited from three hospitals in the greater Houston area, TX, USA. Inclusion and exclusion criteria for participants born at term and preterm and definitions of low and high risk of developmental disorders are provided in Tables SI and SII (published online only). Most participants were African-American (63.0%) with the rest being of Caucasian (20.1%) and Hispanic (15.0%) ethnicity. The participants were predominantly from lower socio-economic groups. There were slightly more females than males. Quality of schooling, socio-economic status, sex, and ethnicity were not different across birth groups.
The reading study participants
Recently we studied the relationships of language, intelligence, and executive function to reading ability and birth status in participants from the original cohort during the 3rd, 5th, and 7th school grades. Participants were clustered into poor, average, and good readers by analysis of the level and growth of phonological word decoding as indexed by the Woodcock–Johnson Test of Achievement Word Attack subtest8 (see Table SIII, published online only). Sixteen participants from the original cohort were eliminated because they demonstrated two or more quantitative skill scores below 85 on the Stanford–Binet intelligence scale (4th edn). Because of attrition, 91 participants from the original group could not be assigned to a reading group. This resulted in a total of 253 participants, 70.3% of the original cohort.
The adolescent sample
To study brain–behavior relationships in a group of adolescents with a wide range of reading abilities, we attempted to recruit five participants for each combination of reading group and birth risk, with an even sex distribution. Reading grouping was not used further; rather, the actual reading-related skills were analysed in the current study. Right-handedness was confirmed by a laterality index as assessed by an Edinburgh Handedness Inventory9 score of more than 50.10 After description of the study, written informed consent was obtained in accordance with our institutional review board Regulations for the Protection of Human Subjects. Thirty-two participants had a reading assessment and diffusion tensor imaging scan. Inattention was calculated as the average continuous performance task inattention score across the 3rd, 5th, and 7th grades. Twenty of the adolescents underwent the Test of Variables of Attention, an alternative continuous performance test, during the reading assessments. The standardized attention score of the Test of Variables of Attention correlated with the inattention score from the continuous performance task (r=0.43; t(18)=2.02; p<0.05). Characteristics of the final sample are given in Table I.
Table I. Characteristics of the adolescents participating in this study
Low risk preterm
High risk preterm
Data are medians (interquartile ranges).
Male:female ratio, n
Gestational age, wks
Phoneme reversal score
Word attack score
Letter–word identification score
Rapid naming composite score
Participants were examined using carefully selected tests to assess targeted reading-related skills. Orthographic lexical and sublexical decoding was measured with the Woodcock–Johnson III Letter–Word Identification subtest and the Woodcock–Johnson Test of Achievement Word Attack subtest respectively. Pure auditory phonological awareness skill and speed of retrieval were measured by the Comprehensive Test of Phonological Processing Phoneme Reversal and Rapid Naming Composite respectively.
Neuroimaging data acquisition and processing
We used a high signal-to-noise ratio whole-brain diffusion tensor imaging protocol at 3.0T that was kept under 7 minutes.11 The diffusion-weighted data were collected axially (inferior-to-superior from the foramen magnum to the vertex) using 44 contiguous 3mm sections that covered the entire brain. The diffusion sensitization or b-factor was 1000s/mm2 and the encoding scheme used 21 uniformly distributed directions. Details of the diffusion tensor image processing and data quality-control measures are provided elsewhere.11 Anatomical T1 and T2 images were also acquired and reviewed by a board-certified radiologist or neurologist. One abnormal scan was excluded.
After diffusion-weighted data preprocessing, which included distortion correction, masking, and isotropic voxel interpolation, the diffusion tensors were constructed and diagonalized as described elsewhere.11 The tensor eigen values were used to obtain the transverse, principal mean diffusivities as well as volumes. Gray and white matter was segmented using a diffusion tensor imaging-based algorithm.12 Fiber tracts were automatically segmented using the statistical parametric mapping toolbox (http://www.fil.ion.ucl.ac.uk/spm) and the international consortium for brain mapping probabilistic templates and atlases as described by Mori et al.13 Measures used in this manuscript include the mean pathway fractional anisotropy and radial diffusivity and the pathway volume normalized by total intracranial volume. For this study we specifically identified and analysed the SLF and the superior and inferior frontal-occipital fasciculi.
A general linear mixed model (‘mixed’ procedure of SAS 9.1; SAS Institute Inc., Cary, NC, USA) was used to investigate the relationships between the fixed effects of birth group, hemisphere, and cognitive skill and the white-matter indices (fractional anisotropy, radial diffusivity, volume). The intercept was modeled as a random effect. The full linear model is as follows: white matter index=constant+birth group+hemisphere+birth group×hemisphere+cognitive-skill+birth group×cognitive-skill+hemisphere×cognitive-skill+birth group×hemisphere×cognitive-skill. Before the full model was calculated, a reduced model that did not include the cognitive skill was calculated to investigate the effect of birth group and hemisphere on white-matter structure. The full model was calculated for each cognitive skill separately.
Models were simplified hierarchically by removal of the highest-order non-significant interaction or fixed effect if all interactions containing it were eliminated. The model was simplified until an interaction was significant, in which case all effects contained within the interaction were retained, or all effects in the model were significant. This statistical simplification procedure is commonly used, and was used in our previous behavioral and brain imaging studies.3,14 Because we made multiple comparisons using the same structural measure (i.e. five behavioral performance variables) we set alpha to 0.01 for the full analysis.15 However, an alpha of 0.05 was used for the initial reduced analysis that did not include cognitive skills. To better understand the relationships between cognitive skills and microstructure when interactions were present, we used Pearson’s correlation statistics for each group separately. This was done for descriptive purposes with the caveat that correlation analysis was less sensitive than the interaction, because of the reduced sample size relative to the total sample. When the hemisphere effect or its interaction was not significant, the microstructure measurements were averaged across the hemisphere before graphing or performing correlations to mitigate the effect of repeated observation.
Superior longitudinal fasciculus
Neither fractional anisotropy nor radial diffusivity was different across birth groups. Radial diffusivity was larger in the right than in the left hemisphere (F(1,30)=19.69, p<0.01) (Fig. 1). Fractional anisotropy was not different across hemispheres. Volume was not different across birth groups or hemispheres.
Superior frontal-occipital fasciculus
The relationship between radial diffusivity and birth group was different across hemispheres (F(2,28)=4.54, p=0.02). Radial diffusivity became increasingly different across hemispheres as birth risk increased from term to preterm low risk to preterm high risk. Fractional anisotropy was not influenced by either birth group or hemisphere (Fig. 1). Volume was greater in the right than in the left hemisphere (F(1,31)=40.97, p<0.01) but was not different across birth groups (Fig. 2).
Inferior frontal-occipital fasciculus
The relationship between fractional anisotropy and birth group was different across hemispheres (F(2,29)=5.09, p=0.01). As seen in Fig. 1, fractional anisotropy was higher in the left than in the right hemisphere for adolescents born at term, equal across hemispheres for adolescents born preterm at low risk, and greater in the right than in the left hemisphere for adolescents born preterm at high risk. This difference across hemispheres was driven by changes in fractional anisotropy for the left but not the right hemisphere. Radial diffusivity was significantly smaller in the left than in the right hemisphere (F(1,31)=10.35, p<0.01; Fig. 1). Volume was not different across birth groups or hemispheres.
Performance-associated effects that differ across hemispheres
Superior longitudinal fasciculus
The relationships between fractional anisotropy and letter–word identification (F(1,30)=20.03, p<0.01), phoneme reversal (F(1,30)=6.69, p=0.01), and inattentiveness (F(1,30)=8.68, p<0.01) were different across hemispheres. The relationships between fractional anisotropy and performance on letter–word identification (r(30)=−0.37, p<0.05) and phoneme reversal (r(30)=−0.34, p=0.05) were significant in the left, but not the right, hemisphere (Fig. S1, published online only), whereas the relationship between fractional anisotropy and inattentiveness was significant in the right hemisphere (r(30)=−0.38, p<0.05), but not the left hemisphere (Fig. S1). Fractional anisotropy decreased as task performance increased for the reading-related tasks and decreased as performance decreased for the attention task.
The relationships between radial diffusivity and performance on letter–word identification (F(1,30)=6.39, p=0.01) and phoneme reversal (F(1,30)=7.82, p<0.01) were different across hemispheres, with the relationships being significant or near significant between left, but not right, hemisphere radial diffusivity and performance on the letter–word identification (r(30)=0.31, p<0.10) and phoneme reversal (r(30)=0.40, p<0.05; Fig. S1). Radial diffusivity increased as task performance increased for the reading-related tasks.
Performance-associated effects that differ across birth groups
Superior longitudinal fasciculus
The relationships between volume and phoneme reversal (F(2,32)=4.90, p=0.01) and inattention (F(2,32)=5.32, p<0.01) were different across birth groups. Reduced SLF volume was associated with poorer performance on phoneme reversal for all adolescents born preterm (r(17)=0.58, p<0.01). This relationship was significant for adolescents born preterm at high risk (r(9)=0.70, p<0.02) and near significant for adolescents born preterm at low risk (r(6)=0.63, p<0.10), whereas there was no relationship for adolescents born at term (Fig. 2a). Reduced SLF volume was associated with higher inattentiveness for all adolescents born preterm (r(17)=−0.69, p<0.01), with this relationship being significant for adolescents born preterm at high risk (r(9)=−0.77, p<0.01) but not those at low risk. Adolescents born at term demonstrated the opposite relationship: increased SLF volume was associated with greater inattentiveness (r(11)=0.59, p<0.05; Fig. 2b).
In this study we examined the brain–behavior relationships between reading-related and attention skills and structural characteristics of white-matter tracts that connect the frontal lobes with posterior areas of the brain (i.e. parietal, temporal, and occipital lobes) in adolescents born at term and those born preterm at low risk and high risk of developmental disorders. Specifically, we examined the microstructure and macrostructure of the SLF and superior and inferior frontal-occipital fasciculi. Our findings suggest that the microstructure of the SLF may be related to reading-related and attention skills, with this relationship demonstrating different lateralization depending on the skill. The two reading-related tasks, letter-word identification and phoneme reversal, were related to left-hemisphere SLF microstructure, whereas the attention skill was related to right-hemisphere SLF microstructure. This relationship was not different across birth groups, suggesting that changes in SLF microstructure may be a general neuropathological marker of cognitive dysfunction. In contrast, the relationship between SLF macrostructure and cognitive function was different across birth groups. Specifically, poorer performance on reading-related and attention tasks was related to a reduction in SLF volume in adolescents born preterm but not in adolescents born at term. This suggests that common microstructural changes seen in adolescents born preterm and at term may arise from different developmental processes. A relationship between cognitive skills and the frontal-occipital fasciculi structure was not found in this study. Unlike the SLF, the microstructure of the frontal-occipital fasciculi was influenced by birth risk, possibly adding variability to the microstructure measures, resulting in poor brain–behavior correlations.
Animal, genetic, and imaging studies have converged on abnormalities in brain connectivity underlying reading disability.16 Genes associated with reading disability appear to be linked to neural migration and axonal guidance.17 Both functional and anatomic neuroimaging studies suggest abnormal connectivity between language areas in individuals with reading disability.7,18–20 Our data are consistent with the notion that the SLF is important in reading. This pathway connects cortical regions hypothesized to be involved in reading,6 and diffusion tensor imaging studies have implicated white-matter disorganization of the SLF in reading disability.7,18 The current study suggests that the microstructure of the SLF is similarly disrupted in adolescents born at term and preterm.
We have demonstrated that better performance on certain reading-related tasks was associated with a lower fractional anisotropy and higher radial diffusivity in the SLF. These patterns of microstructure changes are consistent with previous studies of the corpus callosum14,19 and the temporoparietal white matter.20 Dougherty et al.19 suggested that this effect could be due to differences in the type and size of the axons. Specifically, a higher fractional anisotropy and lower radial diffusivity could be consistent with a decrease in the number of larger-diameter fast axons. Such a notion would certainly be consistent with the neuropathological loss of magnocellular neurons and increased latencies of evoked responses documented in individuals with reading disability.21,22 However, other interpretations are equally probable. For example, a lower fractional anisotropy, especially calculated on the whole-pathway volume, might indicate a simpler white-matter pathway with less crossing fibers. Further research will be needed to differentiate these multiple possibilities.
The data from this study suggest that microstructure of the right-hemisphere SLF is related to inattention. The relationship between poor connectivity linking the posterior and anterior brain and poor performance on the continuous performance task is consistent with functional imaging studies. Functional imaging has confirmed that frontal brain areas, including the inferior, middle superior, and orbital areas, participate in an extensive brain-wide neural network during the continuous performance task.23 The fact that our previous study demonstrated an association between reading and executive function in children born preterm at high risk is consistent with neuroimaging studies of preterm children that suggest a general disconnection between the anterior and posterior brain region. Children born preterm demonstrate abnormal microstructure in white-matter pathways connecting frontal and parietal, temporal, and occipital areas,5 and reduced modulation of frontal areas during a passive auditory language listening task.24 Indeed, individuals born preterm may find it difficult to recruit frontal brain areas during reading and executive-function tasks. For example, functional imaging studies have suggested that children born preterm use alternative strategies to process language that may not require frontal brain areas, such as over-reliance on semantic information.24
Volume of the SLF was found to be related to performance on a reading-related and attention task for adolescents born preterm, but not at term. This finding is consistent with known neuropathology associated with preterm birth. Indeed, neuropathological studies have suggested that white-matter volume loss in infants born preterm at very low birthweight is associated with axonal or oligodendrocyte loss.25 Diffusion tensor imaging studies have demonstrated patterns of microstructure abnormalities consistent with axonal and oligodendrocyte loss in widespread white-matter areas.5 This suggests that the neuropathology associated with cognitive dysfunction in adolescents born preterm is much more severe than the neuropathology associated with equivalent cognitive dysfunction in adolescents born at term. This also suggests that clinical scans may not be sensitive enough to detect these subtle changes, and quantitative methods may be useful to predict cognitive dysfunction in adolescents born preterm.
In this article we focused on white-matter correlates of cognitive dysfunction in a population of adolescents born preterm and at term. We found both similarities and differences between adolescents born preterm and at term in the association between white-matter structure and reading and attention. Understanding the structural changes that underlie poor reading in adolescents born preterm should help delineate how to better target remediation and medical therapies, both at birth and throughout life. Identifying the neuroimaging signature of white-matter damage associated with cognitive dysfunction in adolescents born preterm may provide a biomarker for risk of cognitive dysfunction.
This study presents a starting point for larger studies of additional important factors that might influence brain development, such as sex. Future studies should also attempt to stratify across sex to provide a more balanced sample. This study and our previous studies have demonstrated that spoken and written language and executive function are important cognitive domains to study in children born preterm. More extensive achievement and neuropsychological batteries performed at the time of magnetic resonance imaging acquisition would be optimal. From our analysis of this study, we have estimated the necessary sample size to achieve this goal. We used linear regression to estimate the r2 for the mixed-model analysis outlined above, which ranged from 0.10 to 0.39. This corresponded to an effect size of 0.11 to 0.64. Given this effect size and a linear model that includes sex, birth risk, and cognitive performance with an alpha of 0.01 and power of 95%, a sample size from 66 to 311 would be needed, depending on the structural measure being analysed.
What this paper adds
• Adolescents born at term and preterm have similar white-matter microstructure abnormalities related to cognitive deficits.
• Adolescents born preterm, but not those born at term, have white-matter macrostructural abnormalities despite normal clinical neuroimaging.
• Macrostructural abnormalities in adolescents born preterm are related to cognitive deficits.
This study was supported by grant NS046565 to Dr Richard E Frye, NINDS R01-NS052505-04 to Dr Khader M Hasan, and HD25128 to Dr Susan Landry.