Main

Preterm birth accounts for 11% of all live births worldwide (1). In addition to neurodevelopmental difficulties, preterm birth is associated with reduced total and regional brain volumes, including white and gray matter, cerebellum, hippocampus, caudate, and corpus callosum (2, 3). Considering increasing survival rates and a consequential increased risk of adverse neurodevelopmental outcomes (4), it is pertinent to determine modifiable risk factors to improve the health and welfare of preterm babies.

Nutrition, including polyunsaturated fatty acids (PUFAs), has a central role in brain development (5, 6). Most brain PUFAs are accumulated between the last trimester of gestation and 2 years of life, during the phase of rapid brain growth (7, 8, 9). Preterm birth interrupts the transfer of placental PUFAs to the fetus, limiting preterm neonatal PUFA stores, and interfering with brain development (6, 7, 8).

Magnetic resonance imaging (MRI) studies investigating the effects of nutrition, including PUFAs, on brain volumes of people born preterm are limited and results are variable (5, 10, 11, 12). Limitations of prior studies include variability in nutritional interventions and lack of postnatal MRI scanning. To our knowledge, no studies have explored the effects of PUFAs on brain volumes of preterm neonates using postnatal MRI scans. Further, of the studies that have utilized near-term MRI scans to examine the relationship between brain volumes and neurodevelopmental outcome in the preterm population, all demonstrate that smaller brain volumes—total and regional—predict worse developmental outcomes during childhood (13, 14, 15, 16).

We recently reported that, during the postnatal period, higher ω-3 and lower ω-6 fatty acid levels are associated with improved microstructural brain development and improved developmental outcomes at preschool age in preterm infants (17). Here we will expand upon our previous findings (17) by assessing the association between postnatal PUFA levels and preterm macrostructural brain development—brain tissue volumes acquired using serial postnatal MRI scans—and also investigate the relationship between brain tissue volumes and developmental outcomes at preschool age in infants born preterm. All things considered (13, 14, 15, 16, 17), it is hypothesized that, during the postnatal period, higher ω-3 and lower ω-6 fatty acid levels will be associated with larger brain tissue volumes, which will predict improved developmental outcomes at 30–36 months corrected gestational age (GA).

Methods

Study Population

Sixty preterm infants (<32 weeks GA) born between March 2010 and November 2011 admitted to the Neonatal Intensive Care Unit at either UCSF Benioff Children’s Hospital at the University of California San Francisco or University of British Columbia affiliated BC Women’s Hospital were studied. Infants were excluded if they had a congenital malformation or syndrome, congenital infection, or were too unstable to transport to the MRI scanner. Clinical history was obtained from patient charts. This study was approved by both institutions’ research ethics boards. Parental consent was obtained for each participating infant. Patient demographics and detailed feeding practices at both study sites were previously described (17).

MRI Studies

MRI scans were performed soon after birth as clinically stable and again at near-term age, using 1.5-T MRI scanners (General Electric Sigma, GE Medical Systems, Milwaukee, WI or Siemens Avanto, Siemens Medical Solutions, Malvern, PA). Conventional MRI sequences were acquired at both study sites (18). Blinded to patient medical history, one neuroradiologist at each study site (A.J.B., K.J.P.) reviewed and graded all scans for severity of intraventricular hemorrhage (IVH) and white matter injury. The highest severity of injury score from both scans was used to delineate the patient’s severity of injury (17).

Brain Tissue Volumes

Each subject’s T1-weighted MRI scan was automatically segmented into tissue classes using a spatio-temporal atlas-driven expectation maximization algorithm (19). This incorporated a bias estimation step and used a 4D atlas built from a database of 32 manually delineated MRI scans. This approach uses patch-based priors to augment the age-specific atlas to better account for abnormal anatomy not captured by the atlas training data. The system has been carefully evaluated using a similar cohort of T1-weighted scans with manually marked tissues as a reference. The Dice Similarity Coefficient was used to quantify the accuracy of the automatic segmentation with an average of 0.883 across tissue classes (19). Tissue classes included: cortical gray matter, white matter, deep gray matter, cerebellum, brainstem, and ventricular cerebrospinal fluid (Figure 1). After labeling all MRI voxels, the volume of each tissue class was calculated for each subject scan.

Figure 1
figure 1

Brain tissue volumes at near-birth and near-term age. Segmentations of cortical gray matter (peach), white matter (light blue), deep gray matter (fuchsia), cerebellum (dark blue), brainstem (green), and ventricular cerebrospinal fluid (yellow) superimposed on 3D T1-weighted magnetic resonance images obtained at (a) near-birth and (b) near-term age. Images are shown in axial, coronal, and sagittal planes. R, right.

Full size image

Blood Samples

For each patient, within 7 days of MRI, a 1 ml blood sample was collected at both near-birth and near-term MRI scans, for a total of 2 blood samples. Red blood cell membrane PUFA levels (percent total fatty acids), including ω-3 fatty acids docosahexaenoic acid (DHA; 22:6ω3), eicosapentaenoic acid (EPA; 20:5ω3), ω-6 fatty acids arachidonic acid (ARA; 20:4ω6), and linoleic acid (LA; 18:2ω6), were determined by gas chromatography–flame ionization detector (20).

Neurodevelopmental Outcome

At 30–36 months corrected age, infants were assessed with the Bayley Scales of Infant and Toddler Development, Third Edition (Bayley-3). Blinded to the infant’s neonatal course, a developmental psychologist or physiotherapist determined the infant’s language, motor, and cognitive composite scores according to the infant’s corrected age.

Statistical Analysis

Statistical analyses were performed using R, version 3.1.1 (The R Foundation for Statistical Computing 2014, Vienna, Austria). A cube-root transformation for brain tissue volumes was used in the analyses. Mixed regression models were used to assess the relationship between fatty acid levels from the near-birth blood sample and brain tissue volumes from both scans to account for repeated measures. Linear regression analysis was used to assess the relationship between fatty acid levels and brain tissue volumes acquired at near-term age. Results for the mixed regression analyses and the linear regression analyses were first only adjusted for age at MRI scan (known clinical risk factor for brain injury) and then further adjusted for measures of brain injury, including IVH severity, white matter injury severity, and cerebellar hemorrhage (for cerebellar volumes only), and for known clinical risk factors for brain injury, including duration of intubation, sepsis, and patent ductus arteriosus. All analyses were also adjusted for study site to account for site-specific variations. Linear regression analysis was used to assess the relationship between near-term brain tissue volumes and Bayley-3 composite scores at 30–36 months corrected age. Results were adjusted for age at near-term MRI scan. The above analyses were then repeated using the same study cohort but excluding subjects with IVH. Associations were considered significant at P⩽0.050.

Results

Study Cohort

The study cohort comprises 60 preterm neonates <32 weeks GA enrolled at two study sites. All subjects had an MRI acquired as soon after birth as clinically stable (mean 31.54±2.26 week GA) and associated blood sampling within 7 days. Among the 60 subjects, 44 (73%) subjects also had an MRI at near-term age (mean 37.71±2.98 week GA), along with associated blood sampling (Table 1). Details regarding patient demographics and fatty acid composition of blood samples were previously reported (17).

Table 1 Patient cohort characteristics at near-birth and near-term age
Full size table

Near-birth PUFA Levels and Brain Tissue Volumes

Of the 60 subjects, 59 (98%) subjects’ near-birth MRI scans had adequate scan quality to undergo automatic segmentation of tissue classes. Of the 44 subjects who had an MRI at near-term age, 39 (89%) subjects’ near-term MRI scans were of adequate quality to undergo automatic segmentation. Near-birth PUFA levels were correlated with brain tissue volumes from both near-birth and near-term MRI scans. Adjusting for age at each MRI scan, mixed effects analysis revealed positive associations for DHA with cortical gray matter (β=3.285 cm3/1% PUFA, P=0.008), deep gray matter (β=0.729 cm3/1% PUFA, P=0.003), cerebellar (β=0.650 cm3/1% PUFA, P=0.002), and brainstem (β=0.189 cm3/1% PUFA, P=0.005) volumes; for ARA with deep gray matter (β=0.263 cm3/1% PUFA, P=0.040) and cerebellar (β=0.226 cm3/1% PUFA, P=0.045) volumes; negative associations for EPA with cortical gray matter (β=−16.704 cm3/1% PUFA, P=0.034), white matter (β=−25.922 cm3/1% PUFA, P=0.021), deep gray matter (β=−3.874 cm3/1% PUFA, P=0.012), and cerebellar (β=−3.283 cm3/1% PUFA, P=0.018) volumes; and for LA with cortical gray matter (β=−1.404 cm3/1% PUFA, P=0.020), white matter (β=−2.488 cm3/1% PUFA, P=0.003), deep gray matter (β=−0.384 cm3/1% PUFA, P=0.001), cerebellar (β=−0.246 cm3/1% PUFA, P=0.022), and brainstem (β=−0.074 cm3/1% PUFA, P=0.029) volumes (Table 2, Model 1). After further adjusting for IVH severity, white matter injury severity, duration of intubation, sepsis, patent ductus arteriosus, and cerebellar hemorrhage (for cerebellar volumes only), only DHA remained positively associated with cortical gray matter (β=3.366 cm3/1% PUFA, P=0.022), deep gray matter (β=0.721 cm3/1% PUFA, P=0.007), and brainstem (β=0.168 cm3/1% PUFA, P=0.042) volumes and only LA remained negatively associated with white matter (β=−1.964 cm3/1% PUFA, P=0.050) volume (Table 2, Model 2). One percent increase in DHA was associated with a 3.366 cm3 increase in cortical gray matter volume, 0.721 cm3 increase in deep gray matter volume, and 0.168 cm3 increase in brainstem volume at near-term age. Further, a 1% increase in LA was associated with a 1.964 cm3 decrease in white matter volume at near-term age. No associations were found for any studied PUFA levels and ventricular cerebrospinal fluid volumes (P>0.100).

Table 2 Mixed effects analyses comparing red blood cell fatty acid levels (percentage of total fatty acid) from first blood sample and brain tissue volumes (cm3) from two MRI scans acquired as soon after birth as clinically stable and at near-term age
Full size table

Considering previously reported results in the same cohort of an association between IVH and DHA and the potential impact of IVH on brain growth (17), subanalysis was performed excluding infants with IVH. Adjusting for age at each MRI scan, white matter injury severity, duration of intubation, sepsis, patent ductus arteriosus, and cerebellar hemorrhage (for cerebellar volumes only), mixed effects analysis revealed positive associations for only DHA with cortical gray matter (β=3.598 cm3/1% PUFA, P=0.028) and deep gray matter (β=0.975 cm3/1% PUFA, P=0.002) volumes. After excluding infants with IVH, DHA was no longer associated with brainstem (β=0.155 cm3/1% PUFA, P=0.131) volume, and LA was no longer associated with white matter (β=−0.143 cm3/1% PUFA, P=0.916) volume. As in prior analyses, no associations were found for EPA (P>0.200) or ARA (P>0.400) with any studied brain tissue volumes.

Near-term PUFA Levels and Near-Term Brain Tissue Volumes

Near-term PUFA levels were correlated with brain tissue volumes from near-term MRI scans. Adjusting for age at near-term scan, linear regression analysis revealed positive associations for DHA with deep gray matter (β=0.689 cm3/1% PUFA, P=0.046), cerebellar (β=0.908 cm3/1% PUFA, P=0.021), and brainstem (β=0.226 cm3/1% PUFA, P=0.013) volumes and for ARA with deep gray matter (β=0.283 cm3/1% PUFA, P=0.039) volume (Table 3, Model 1). After further adjusting for IVH severity, white matter injury severity, duration of intubation, sepsis, patent ductus arteriosus, and cerebellar hemorrhage (for cerebellar volumes only), only DHA remained positively associated with deep gray matter (β=0.822 cm3/1% PUFA, P=0.032), cerebellar (β=0.964 cm3/1% PUFA, P=0.028), and brainstem (β=0.245 cm3/1% PUFA, P=0.021) volumes (Table 3, Model 2). No associations were found for EPA (P>0.100) or LA (P>0.100) with any studied brain tissue volumes or for any studied PUFA levels and white matter (P>0.100) or ventricular cerebrospinal fluid (P>0.100) volumes. One percent increase in DHA was associated with a 0.822 cm3 larger deep gray matter volume, 0.964 cm3 larger cerebellar volume, and 0.245 cm3 larger brainstem volume at near-term age.

Table 3 Linear regression analyses comparing red blood cell fatty acid levels (percentage of total fatty acid) from second blood sample and brain tissue volumes (cm3) from MRI scan acquired at near-term age
Full size table

Excluding infants with IVH and adjusting for measures of brain injury and known clinical risk factors for brain injury did not meaningfully change the positive associations found for DHA with deep gray matter (β=0.852 cm3/1% PUFA, P=0.032), cerebellar (β=0.768 cm3/1% PUFA, P=0.033), and brainstem (β=0.257 cm3/1% PUFA, P=0.007) volumes. However, after excluding infants with IVH, LA became positively associated with cortical gray matter (β=3.472 cm3/1% PUFA, P=0.037) volume. As in prior analyses, no associations were found for EPA (P>0.070) or ARA (P>0.090) with any studied brain tissue volumes.

Near-term Brain Tissue Volumes and Developmental Outcome

Of the 60 subjects, 45 (75%) underwent developmental assessments at a mean corrected age of 32.8±3.6 months. Means for Bayley-3 composite scores were 101.0±17.8 for language, 98.1±14.3 for motor, and 103.6±14.4 for cognitive. Of the 39 subjects who had near-term MRI scans with adequate quality to undergo automatic segmentation of tissue classes, 31 (79%) underwent developmental assessments at 30–36 months corrected age. Near-term brain tissue volumes were correlated with Bayley-3 composite scores at 30–36 months corrected age. Adjusting for age at near-term scan, linear regression analysis revealed positive associations for cortical gray matter (β=0.546 pt/1 cm3, P<0.001), deep gray matter (β=3.379 pt/1 cm3, P=0.004), cerebellar (β=3.480 pt/1 cm3, P=0.001), and brainstem (β=10.853 pt/1 cm3, P=0.033) volumes with language scores and for cerebellar (β=2.073 pt/1 cm3, P=0.016) and brainstem (β=8.444 pt/1 cm3, P=0.030) volumes with motor scores at 30–36 months corrected age (Table 4). No associations were found for ventricular cerebrospinal fluid volume and Bayley-3 composite score in any domain (P>0.400) or for any studied brain tissue volume with cognitive scores (P>0.080) at 30–36 months corrected age. A 1 cm3 larger cortical gray matter volume at near-term age was associated with 0.5 point higher language scores at 30–36 months corrected age. A 1 cm3 larger near-term deep gray matter volume was associated with 3.4 point higher language scores. Further, a 1 cm3 larger cerebellar volume at near-term age was associated with 3.5 point higher language scores and 2.1 point higher motor scores. Lastly, a 1 cm3 larger near-term brainstem volume was associated with 10.9 point higher language scores and 8.4 point higher motor scores.

Table 4 Linear regression analysis comparing brain tissue volumes (cm3) from MRI scan acquired at near-term age and Bayley-3 language, motor, and cognitive scores at 30–36 months corrected age
Full size table

Excluding infants with IVH and adjusting for age at each MRI scan did not meaningfully change the positive associations found for cortical gray matter (β=0.430 pt/1 cm3, P=0.004), deep gray matter (β=2.414 pt/1 cm3, P=0.049), and cerebellar (β=3.053 pt/1 cm3, P=0.011) volumes with language scores. However, brainstem (β=7.453 pt/1 cm3, P=0.199) volume was no longer associated with language scores, and cerebellar (β=1.097 pt/1 cm3, P=0.184) and brainstem (β=3.966 pt/1 cm3, P=0.298) volumes were no longer associated with motor scores. As in prior analyses, no associations were found for any studied brain tissue volume with cognitive scores (P>0.300) at 30–36 months corrected age.

Discussion

Utilizing serial blood samples, neuroimaging, and developmental assessments in preterm neonates, we demonstrate opposing associations with respect to ω-3 and ω-6 fatty acid levels and their effect on brain tissue volumes during the neonatal period, showing larger neonatal brain tissue volumes associated with improved developmental outcomes at preschool age. Higher near-birth DHA levels are associated with larger cortical and deep gray matter and brainstem volumes and higher near-term DHA levels are associated with larger deep gray matter, cerebellar, and brainstem volumes at near-term age. Moreover, we demonstrate an association between higher near-birth LA levels and decreased near-term white matter volumes. Larger cerebellar and brainstem volumes at near-term age predict improved Bayley-3 language and motor scores at 30–36 months corrected age. Further, larger near-term cortical and deep gray matter volumes predict improved Bayley-3 language scores at 30–36 months corrected age.

Considering that DHA accretion relative to brain weight is highest during fetal development and early infancy (21), and its preferential placental transfer to the fetus during the third trimester, it is not surprising that, as the most abundant ω-3 fatty acid in the mammalian brain, DHA is essential to brain development (5, 7, 21, 22). This study finds that, for only DHA levels, both near-birth and near-term blood samples are associated with increased brain tissue volumes at near-term age, albeit differences in affected brain regions. Increased DHA levels at both near-birth and near-term time points are associated with larger deep gray matter and brainstem volumes, suggesting a larger sensitive time window for DHA to positively affect these regions. Higher DHA levels only at the near-birth time point are associated with larger cortical gray matter volumes, suggesting a shorter and earlier time period for DHA’s influence, and providing evidence that near-birth DHA levels predict near-term cortical gray matter volumes. Further, greater DHA levels only at the near-term time point are correlated with larger cerebellar volume, suggesting a shorter window for DHA to impact the cerebellum. The mechanisms—neuroprotection or growth enhancement—by which DHA is affecting individual brain tissue regions have yet to be elucidated. Although DHA is present throughout the brain, the cortex has the highest percentage of DHA, whereas the medulla has the lowest (23). DHA is also present in high quantities in the phospholipids of brain gray matter (9, 24, 25). Of note, whole-blood DHA levels are lower in preterm than in full-term newborns (22). However, no studies have previously reported the effects of DHA on volumes of specific brain regions.

Initially, it may seem as though the current DHA results conflict with previously reported diffusion tensor imaging results in the same cohort, which showed that higher DHA was associated with diffusion tensor imaging measures in regions of interest located in white matter tracts (17), whereas the current study found no association between DHA and white matter volumes. However, as prior diffusion tensor imaging results showed higher DHA associated with decreased mean diffusivity and no association with fractional anisotropy, it was suggested that improved microstructural development was independent of myelination (17). This supports the current findings that DHA levels were not associated with white matter volumes.

In contrast to DHA, levels of the ω-6 fatty acid LA remain relatively stable during late gestation (26), are lower in fetal than maternal plasma and erythrocyte phospholipids (27, 28), and LA transfer to the fetus by the placenta during pregnancy is restricted (29). This study determines that LA levels only at the near-birth time point are associated with brain tissue volumes. Higher near-birth LA levels are correlated with decreased near-term white matter volumes, suggesting an earlier and shorter sensitive window for LA to negatively affect white matter, and providing evidence that near-birth LA levels predict near-term white matter volumes. Previous findings suggest that low placental transfer of LA is a favorable biological mechanism that ultimately facilitates DHA accretion in the fetus (29). LA and α-LA compete for the same desaturation enzymes that ultimately convert these essential fatty acids to ARA and DHA, respectively. In the developing brain, although α-LA is the preferred substrate for these desaturases, elevated levels of LA can result in inhibition of DHA production, potentially posing developmental problems during the perinatal period and into childhood (30). This supports our finding that DHA and LA have opposing associations with brain tissue volumes, which is strengthened by previous findings that revealed that higher DHA and lower LA levels were associated with improved microstructural development (17). Future studies need to decipher details behind the negative association between near-birth LA levels and near-term white matter volume. Animal studies have shown that during the neonatal period the difference in LA concentration across brain regions is not very apparent (23). Studies have also shown that LA levels in preterm newborns are altered when compared with full-term infants (22). No previous investigations have described a relationship between LA and regional brain volumes.

We previously determined that higher DHA and lower LA levels during the postnatal period were associated with improved developmental outcomes in preterm newborns (17). Higher near-birth DHA levels were associated with higher language and motor scores and lower near-birth LA levels were associated with higher motor and cognitive scores on the Bayley-3 (17). This study reveals that improved macrostructural brain development—larger brain tissue volumes—at near-term age is also associated with improved developmental outcomes at 30–36 months corrected age, supporting and expanding upon previous findings. Larger cortical and deep gray matter, cerebellar, and brainstem volumes at near-term age are associated with improved language scores, suggesting that language development in the preterm population may be dependent on postnatal volumes of multiple brain regions, and disruption of typical regional brain growth by preterm birth may lead to delayed and/or impaired language development detected at preschool age. Further, larger cerebellar and brainstem volumes are also associated with improved motor scores, suggesting that preterm motor development at preschool age can be predicted by postnatal cerebellar and brainstem volumes and that the cerebellum and brainstem have a more global impact on the development of preterm born children. Of the few studies that employed near-term MRI scans to investigate the relationship between brain volumes and neurodevelopmental outcome in preterm newborns, all illustrate that smaller global and regional brain tissue volumes—cerebrum, frontal lobes, cortical gray matter, basal ganglia, thalami, cerebellum—at near-term age predict worse neurodevelopmental outcomes at preschool age (13, 15, 16), supporting our findings.

The impact of neonatal feeding practices on PUFA levels for this study cohort was previously reported (17). Reported results suggest that feeding regimens currently employed in the Neonatal Intensive Care Unit have no significant effect on postnatal DHA levels for the study cohort and thus should not impact significant associations of DHA with brain tissue volumes reported here. However, higher near-birth LA levels were found to be associated with increased duration of intravenous nutrition (β=1.96, P=0.001) and lipid supplementation (β=1.88, P=0.001). This finding can be expected as the intravenous fat emulsion contains LA and suggests that the measures of estimated nutritional exposures should be able to identify differences in feeding.

As the sickest newborns were harder to feed, it could be argued that measures of poor brain growth—smaller brain tissue volumes—are related to disease rather than PUFA levels. This is unlikely to be the case for findings presented here because the analyses were controlled for measures of brain injury, including IVH severity, white matter injury severity, and cerebellar hemorrhage (for cerebellar volumes only) and for known clinical risk factors for brain injury, including age at MRI scan, duration of intubation, sepsis, and patent ductus arteriosus. In addition, as previously reported results in the same cohort illustrated an association between IVH and DHA and the potential impact of IVH on brain growth (17), analyses were repeated excluding infants with IVH. In general, after adjusting for potential confounders, the associations between the studied PUFAs (i.e., DHA, EPA, ARA, and LA) and brain tissue volumes did not meaningfully change. This strengthens the likelihood that the associations found between brain tissue volumes and PUFA levels are real. However, the influence of disease on the brain tissue volumes cannot be ignored entirely, as there may be residual confounding.

Of note, red blood cell membrane PUFA level is a cell marker depicting a cell membrane synthesized weeks earlier (6), thus near-birth PUFA levels presented here may not only represent early postnatal levels but may also reflect intrauterine levels. As this study finds that DHA levels from near-birth blood samples are significantly associated with larger brain regions, DHA may be important for growth and/or neuroprotection of certain brain regions throughout the perinatal period as opposed to only the postnatal period. This evidence along with other studies which have determined that women with higher plasma DHA during pregnancy give birth to newborns with higher DHA (24, 25) provide support for therapeutic intervention with DHA to help improve region-specific brain growth in at-risk pregnant women.

Although the size of study cohort is limited, adjustments for analyses were made for potential confounders selected a priori based on previous studies (17) and the number of potential confounders controlled for were limited due to sample size. However, many associations between postnatal PUFA levels and brain tissue volumes and between brain tissue volumes and developmental outcomes at 30–36 months corrected age remained across multiple analyses, providing evidence that these study findings are genuine.

This study demonstrates that the associations between PUFAs and the brain during development are not homogenous. We find that ω-3 and ω-6 fatty acids have opposing associations with brain tissue volumes, with higher ω-3 and lower ω-6 fatty acids levels correlated with larger volumes. Further, the correlations between PUFAs and brain growth were found to be time dependent and region specific. The associations between DHA and brain tissue volumes are variable and occur throughout the perinatal period, whereas only near-birth LA levels predict only white matter volumes. In accordance with previous literature (13, 15, 16, 17), albeit differences in affected brain regions and associated outcomes, this study provides further evidence that improved macrostructural brain development during the neonatal period can predict improved developmental outcomes during the preschool period. Future studies need to investigate therapeutic windows and thresholds for individual PUFAs and their associations with specific brain regions during development and associations with long-term developmental outcomes. Our current findings suggest that it may prove beneficial for therapeutic interventions to target both at-risk expectant mothers and preterm neonates.