Altered microstructure of the splenium of corpus callosum is associated with neurodevelopmental impairment in preterm infants with necrotizing enterocolitis

This was a prospective observational cohort study involving postnatal follow-up of preterm infants admitted to the Hanyang Inclusive Clinic for Developmental Disorders in Hanyang University, College of Medicine. Informed consent was obtained from parents of all the children included in this study. The inclusion criteria were as follows: [1] preterm birth weight < 1500 g and a diagnosis of NEC stage II or greater [2]; no evidence of congenital malformation [3]; no evidence of intrauterine growth retardation (IUGR) [4]; no evidence of any grade of intraventricular hemorrhage (IVH) and periventricular leukomalacia (PVL) [5]; no evidence of spontaneous intestinal perforation (SIP); and [6] availability of brain magnetic resonance imaging (MRI) at TEA. Demographic and clinical data were prospectively recorded, including maternal information, gestational age (GA), birth weight, delivery type, sex, histology of placenta, antenatal steroid, and Apgar score. Neonatal data including NEC, respiratory distress syndrome (RDS), culture-proven sepsis, hypotension, patent ductus arteriosus (PDA), bronchopulmonary dysplasia (BPD), retinopathy of prematurity (ROP), and total parenteral nutrition (TPN) duration were analyzed. BPD was diagnosed based on the need for oxygen support at 28 days of age and at 36 weeks of postmenstrual age (PMA) [11]. All trans-fontanelle ultrasound scans were evaluated for IVH and PVL by a single pediatric radiologist. Early trans-fontanelle ultrasound scans were performed within 3 days of birth and at 1 and 3 weeks after birth. At TEA between 36 and 41 weeks of PMA, we performed brain MRI and DTI to evaluate the structural brain network. Lastly, the Bayley Scales of Infant and Toddler Development-III (BSID-III) was performed at 18 months of CA with cognitive, language, and motor composite scores. The BSID-III test is composed of an average of 100 and a standard deviation of 15 in each composite score. The NEC group was defined as preterm infants who were diagnosed with NEC stage II or greater according to Bell’s Modified staging criteria [12]. The control group was defined as preterm infants without evidence of NEC, congenital malformations, IUGR, IVH, PVL, and SIP.

MRI scans were performed for preterm infants at TEA during natural sleep using a 3.0-T MRI scanner (Philips real-time compact magnet 3.0-T MRI system; Achieva 3.0 T X-Series) with a 16-channel SENSE head coil. The T1-weighted images included sagittal and axial T1 spin-echo sequences (400/25/2, repetition time [TR]/echo time [ms]/signal intensity average) and axial T2 turbo spin-echo sequences (3000/100/1). Cushions were placed between the subject and the radiofrequency coil during image acquisition. DTI was performed using a single-shot spin-echo-planar sequence with a SENSE factor of 2 and an echo-planar imaging factor of 51 (TR/TE, 8100/75 ms; matrix size, 112 × 112; field of view, 224 mm; 74 axial sections). The slice orientation was axial with a 2.0-mm thickness and parallel to the anterior-posterior commissure line. Forty to fifty slices covered the entire hemisphere and brainstem. Fifteen directions using an electrostatic gradient model (b = 800) were used for diffusivity measurement. The subjects were well-fed before the scan, and sedative medications were not used.

Brain volumes of preterm infants were measured using an advanced segmentation technique: Morphologically Adaptive Neonatal Tissue Segmentation (MANTiS; http://​developmentalima​gingmcri.​github.​io/​man-tis), which is modified by Statistical Parametric Mapping (SPM) software [13]. It has modifications for neonatal imaging using neonate templates [14]. The pipeline classifies a T2-weighted MRI image of the brain into the following six sub-regions: cortical gray matter, cerebral white matter, cerebellum, subcortical gray matter (including deep nuclear gray matter, the hippocampus, and the amygdala), brainstem, and cerebrospinal fluid. Brain extraction was conducted using the Brain Extraction Tool (BET) in the FMRIB’s Software Library (FSL; http://​www.​fmrib.​ox.​ac.​uk/​). Then, using the “new segmentation tool” SPM 12, the initial tissue was classified with a neonate probability map included in MANTiS. Morphological watershed segmentation and filtering were processed against large ventricles or high-intensity white matter for reliable segmentation. Except for BET, the above processes were automatically executed using the MANTiS pipeline (Fig. 1).

The diffusion-weighted images were processed using the FMRIB’s Diffusion Toolbox (FDT) from FMRIB’s Software Library (Fig. 2). Motion artifacts and eddy current distortions were corrected by normalizing each diffusion-weighted image to a non-diffusion-weighted image (b0) using FMRIB’s linear image registration tool [15, 16]. BET was used to remove non-brain tissues [17]. Every voxel in the diffusion tensor was estimated using least-squares optimization [18]. A scalar map, including three values (λ1, λ2, λ3), fractional anisotropy (FA), and mean diffusivity (MD), was obtained. We used the probabilistic maps of the fiber pathways with a threshold of 0.1 based on the JHU-neonate atlas from Johns Hopkins University [19]. We used established probabilistic maps of fiber pathways that were constructed by DTI tractography and measured the average trace values of brain regions with a probability of more than 10%. The probabilistic map of white matter tracts was overlaid on the JHU-neonate atlas to quantify the FA and MD of the pathway-of-interest related to cognitive, language, and motor function. Regions of interest included the genu of the corpus callosum (gCC), splenium of the corpus callosum (sCC), left corticospinal tract (ltCST), right corticospinal tract (rtCST), left inferior longitudinal fasciculus (ltILF), and right inferior longitudinal fasciculus (rtILF). Furthermore, for a schematic understanding, we selected a single subject for each group and visualized the white matter integrity with threshold of 0.1 using a tractography software tool for MRI Analysis (DSI Studio, http://​dsi-studio.​labsolver.​org/​) (Fig. 3).

Table 1 presents the baseline demographics and clinical characteristics of the study population. Of the 109 preterm infants, we excluded 23 preterm infants and the detailed features were as follows: eight infants were diagnosed with IVH and PVL during NICU admission; six infants were lost to follow-up, nine infants had an insufficient quality of MRI imaging. In total, 86 preterm infants (20 infants with NEC and 66 infants without NEC) were included. Of the 20 infants diagnosed with NEC, three were diagnosed with severe NEC and required surgery. The mean GA of the NEC group and without NEC group was 27.03 and 28.88 weeks, respectively (p = 0.026). The mean PMA at MRI was 37.90 weeks in the NEC group and 36.91 weeks in the without NEC group (p = 0.210). Regarding neonatal characteristics, the NEC group had a significantly longer TPN duration (p = 0.003). Lastly, other neonatal morbidities, including RDS, BPD, PDA, and ROP, did not show significant differences.

Table 2 shows the comparisons of brain volume analysis between the two groups. In every sub-region, preterm infants with NEC showed volume reduction compared to the control group. However, its reduction was not statistically significant after Bonferroni correction. Moreover, its insignificance remained after adjusting for several covariates including sex, PMA at MRI and total intracranial volume.

Representative images of the DTI analysis in this study are shown in Fig. 1. Table 3 shows the network parameters between the two groups. The MD of sCC was significantly higher (p = 0.001) in the NEC group after controlling for covariates. CST resulted in hemispheric differences, showing that MD of ltCST was significantly higher than that in the NEC group (p = 0.001). Lastly, in ILF, network parameters in both hemispheres were not different between preterm infants with and without NEC.

Table 4 presents the neurodevelopmental outcome of preterm infants with and without NEC. Of the 86 preterm infants with MRI analysis, 75 (87%) infants underwent BSID-III in 18 months of CA (19 with NEC and 56 without NEC). Our logistic regression model showed that NEC was a significant risk factor in motor impairment (odds ratio 58.26, 95% confidence internal 7.80–435.12). Looking into the BSID composite scores, the preterm with NEC group had significantly lower BSID-III scores in cognitive (87.00 ± 15.52 vs. 99.04 ± 13.95, p = 0.002), language (82.03 ± 14.03 vs. 92.45 ± 16.42, p = 0.016), and motor (74.42 ± 15.75 vs. 100.36 ± 14.41, p < 0.001) function than those of control group. We performed multivariate linear regression analysis between each composite score of BSID-III and MD of the tract, which was significant in tractography analysis. MD of sCC was significantly negatively associated with language score (p = 0.025) and motor score (p = 0.002). Likewise, in the ltCST, MD was marginally related to motor score (p = 0.058) (Table 5).