ABSTRACT
Aim:
Twin studies allow for the investigation of genetic and environmental influences on human brain development. The generalizability of their findings depends on the developmental similarity between twins and singletons. This study aimed to evaluate the structural and functional differences in a cohort of preterm-born twins and singletons at term-equivalent age.
Materials and Methods:
Eighteen twins and forty-seven singletons were included and scanned at the term-equivalent age. Brain volumes from 3D T1-weighted images, quantitative metrics and structural connectivity from diffusion tensor imaging, and low-frequency brain activity and functional connectivity from resting-state functional MRI (rs-fMRI) were obtained from these neonates.
Results:
We found no significant volumetric differences after multiple comparison correction. The diffusivity values in the cingulum cingular part, cingulate gyrus, lateral fronto-orbital gyrus, gyrus rectus, as well as medial fronto-orbital gyrus were significantly higher in the twin group than in the singleton group. Structural connectivity analysis showed higher transitivity in the twin group compared to the singletons, indicating increased local connectivity. For rs-fMRI, the twin group showed greater fractional amplitude of low-frequency fluctuation (fALFF) values in the salience network and several fronto-temporal regions compared with the singleton group. It is worth noting that we found differences both in structural and functional measurements (MD and fALFF) in the prefrontal and cingulate cortex.
Conclusion:
The structural and functional differences collectively indicated that preterm-born twins may have delayed brain development compared with gestational age-matched singletons at term-equivalent age, which may be related to perinatal-neonatal problems.
Introduction
Knowledge from twin studies allows researchers to understand the contribution of genetic and environmental factors to brain development (1-5). The generalizability of these studies, however, depends on the assumption that brain developmental patterns in twins are comparable to those in singletons. It has been well established that twins have compromised growth in the third trimester starting from about 30 weeks of gestation that may be attributed to certain reasons, such as the different growth pattern, placental size, maternal nutrition, and family care after birth (6-8), leading to potential brain developmental differences between twins and singletons.
Neuroimaging, especially magnetic resonance imaging (MRI), has been shown to be a powerful tool to characterize the structural and functional changes in the developing brain. Brain MRI has been employed to investigate twins and singletons, but not to its full potential. Several existing studies (9-11) focused on morphological differences between twin and singleton brains from 10 months to 30 years of age, however, their conclusions were inconsistent. A recent longitudinal study by Sadeghi et al. (12) compared longitudinal diffusion tensor imaging (DTI) metrics between singletons and twins from birth to 2 years old, and showed that the axial diffusivities in the anterior limb of the internal capsule and anterior corona radiata were significantly higher in twins compared with singletons during early development.
Current MRI studies comparing twins and singletons are still limited and their findings are discrepant, and most studies used a single MRI modality (T1-weighted or DTI). Moreover, all the aforementioned studies focused on comparing twins with term-born singletons, while it is known that twins have an increased risk of preterm delivery (13). The unpaired gestational age makes it difficult to separate the effects of twin birth from premature birth in the observed developmental differences compared with singletons (14). Here, we aim to systematically evaluate whether the MRI findings of twins can be generalized to singleton studies by comparing the preterm-born singleton and twin neonates that were born with equivalent gestational ages with a multi-modal MRI approach, including morphological MRI, DTI, and resting-state functional MRI (rs-fMRI) acquired at term-equivalent age.
Materials and Methods
Preterm-born infants were enrolled at term-equivalent age for MRI scan. Ethical approval was obtained from the Institutional Review Board at the Children’s Hospital of Zhejiang University School of Medicine (2019-11-13). Written informed consent was provided by the parents. Exclusion Criteria included 1) congenital malformation or syndrome; 2) acquired brain injury on MRI; 3) visible artifacts on MRI, or a mean frame-wise displacement (FD) exceeding 0.2 mm for rs-fMRI; 4) psychiatric or neurological family history; 5) pregnancy complications; 6) illicit drug or alcohol use during pregnancy.
Image Acquisition
All neonates received 50 mg/kg oral or enema chloral hydrate 30 minutes before scanning from a radiology nurse who was trained and certified to administer sedation. Ear protectors and physiological monitors were used for protection and monitoring. The scans were performed on a Siemens 1.5T Avanto MRI scanner (Siemens Healthcare, Erlangen, Germany) with a 12-channel Siemens head coil. The multi-modal MRI protocol included the following: 1) 3D sagittal T1-weighted imaging using the MPRAGE sequence with repetition time (TR)=1,910 ms, echo time (TE)=3.01 ms, inversion time=1,100 ms, flip angle=15°, resolution=0.82×0.82×1 mm3, field of view (FOV)=210×210×160 mm3, and acquisition matrix=256×256×160; 2) DTI using a single-shot echo-planar imaging (EPI) sequence with TR=3,800 ms, TE=86 ms, in-plane resolution=1.64×1.64 mm2, FOV=210 × 210 mm2, 25 slices at a slice thickness of 5 mm, 12 gradient directions at a b-value of 750 s/mm2 and 1 non-weighted image (b0), and 4 repetitions; 3) rs-fMRI using a gradient-echo T2*-weighted EPI sequence with TR=2,000 ms, TE=40 ms, in-plane resolution=3.28×3.28 mm2, FOV=210×210 mm2, 24 slices at a slice thickness of 6 mm, bandwidth=200 Hz/pixel, and number of volumes=180.
Image Processing
Image segmentation and structural volumes
The 3D T1-weighted images went through a fully automated segmentation pipeline via the online platform MRICloud (www.mricloud.org) (15), which performed a multi-atlas based segmentation, based on the JHU neonatal multi-atlas (16). Thirty-eight regions of interest (ROIs) were defined, including the gray matter, myelinated/unmyelinated white matter, brain stem, corpus callosum, caudate, putamen, thalamus, and cerebrospinal fluid (CSF), etc. The volumes of the ROIs were obtained and then summed over the two hemispheres (resulting in 18 symmetric ROIs after removing two non-brain-tissue ROIs), assuming negligible laterality in this study.
DTI processing and structural connectivity
All data were manually inspected by a radiologist (T.L.) to exclude diffusion-weighted images (DWI) with noticeable imaging artifacts, followed by intra-subject registration using a 12-parameter affine transformation (17) to correct for head motion. Then, we employed the standard preprocessing steps with denoising (18), Gibb’s ringing removal (19), distortion correction (20), and bias field correction (21). Fractional anisotropy (FA), mean/axial/radial diffusivity (MD/AD/RD) maps were generated from the diffusion tensor using the weighted linear least squares method.
The individual DTI data were transformed to the JHU-neonate single brain DTI atlas for image segmentation (22). The individual mean DWI images were aligned to the atlas DWI image with an affine transformation, followed by histogram matching between the atlas and subject images. Following this, the non-linear transformation was performed with large deformation diffeomorphic metric mapping (23,24), utilizing multi-channel contrasts of the mean DWI, FA, and b0 images (25). The DTI images were further inspected for registration failure, and none of the data showed visible registration errors. After transforming to the atlas space, the individual images were automatically segmented into 126 ROIs, as defined in the JHU-neonate atlas. The FA, MD, AD, and RD values were extracted from the ROIs, and an MD threshold of 2×10-3 mm2/s was used to exclude the CSF voxels. The DTI metrics were then averaged over the hemispheres for statistical analysis.
Tractography was performed on the pre-processed DWIs by a tensor-based probabilistic fiber tracking algorithm (26) in MRtrix3 (www.mrtrix.org). The seed voxels were selected randomly within a whole-brain mask, and the following tracking parameters were used: cut-off of 0.06, step size of 0.16 mm, minimum/maximum length of 8/164 mm. Probabilistic tractography was used in this study as it was shown to yield higher connectome reproducibility than the deterministic method (27,28).
To construct the structural network, all deep white matter and cerebellum ROIs were excluded, leaving 62 ROIs as structural network nodes. It is worth noting that the network nodes included not only cortical and subcortical GM but also subcortical WM because the subcortical WM helped to determine the fibers linked to cortical regions (29). The weakest 1% of the connections, which were considered as spurious streamlines, were discarded. Following this, the streamlines were log-transformed to achieve normality (30). Seven network parameters were calculated using the brain connectivity toolbox (31), including the degree, transitivity, local efficiency, global efficiency, modularity, characteristic path length, and small-worldness.
rs-fMRI Processing
The rs-fMRI data were preprocessed using the data processing assistant for resting-state fMRI (DPARSF, Advanced Edition) (32). First, the first ten time points were removed, followed by slice-timing correction and head motion correction, and those subjects with mean FD exceeding 0.2 mm were excluded (33). Following this, spatial normalization was performed via T1-weighted anatomical images that were registered to the JHU-neonate single brain T1 atlas (22), and all fMRI images were resampled to 3 mm isotropic voxels using SPM8. Next, the normalized images were smoothed with the Full Width at Half Maximum set at 6 mm. Finally, linear drift was removed, and the six rigid head motion parameters, as well as sources of physiological artifact extracted from white matter and CSF masks, were regressed out. In addition, bad time points were scrubbed using a threshold of Jenkinson FD>0.2 mm as well as one volume before and two volumes after (33).
The low-frequency fluctuations were quantified by the amplitude of low-frequency fluctuation (ALFF) and fractional ALFF (fALFF) (34). ALFF was calculated within a specific low-frequency range (0.01-0.1 Hz), then the ratio of the power of the low-frequency band to that of the entire frequency range was calculated as fALFF. Z-transform was performed on both ALLF and fALFF to improve the normality before filtering. Then, the functional connectivity was calculated based on ALFF or fALFF, using the modified atlas as mentioned in 2.3.2 (c), and the correlation coefficient matrices were converted into z map by Fisher’s r-to-z transform to improve normality. Correlation coefficients under 0.2, which were considered to be a negligible correlation, were discarded (35). The same network parameters were calculated as those in the structural network.
Statistical Analysis
For demographic information, categorical data were analyzed using the chi-square test. Shapiro-Wilk’s test was used to analyze the distribution of measurement data. The Student’s t-test was used for normally distributed data, and the Mann-Whitney U test applied to data that did not fulfill the requirements for normality. For the differences in DTI and network metrics between twins and singletons, analysis of covariance (ANCOVA) was performed with a permutation approach in R-Project (36,37). Then the p-values of DTI metrics extracted from multiple ROIs were adjusted with the Bonferroni method. For ALFF and fALFF, a voxel-based analysis of differences between groups was performed with DPABI (32) using a permutation test method followed by Bonferroni correction for multiple testing using threshold-free cluster enhancement (38). The significance level was set at 0.05 for all tests. For all analyses, gender, birth weight, postmenstrual age (PMA) at scan, and Apgar score at 5 minutes after birth were used as covariates.
Results
Demographic and clinical characteristics
In total, eighteen twins and forty-seven singletons were included in this study. For rs-fMRI analysis, eight singletons were excluded as their mean FD exceeded 0.2 mm. Demographic and basic clinical information of the 65 study participants is provided in Table I, and the information of the 57 subjects used for rs-fMRI analysis is listed in Supplementary Table I. No significant group difference was found in terms of the listed demographic and clinical characteristics.
Comparison of the structural volumes
Based on the automated segmentation of the 3D T1-weighted images, the structural volumes of the 18 ROIs were compared between the twins and singletons. The myelinated white matter and thalamus in the singleton group demonstrated higher volumes than the twin group (p=0.034 and p=0.012, respectively before correction), but no significant difference was found after multiple comparison correction. The structural volumes and the statistical test results are listed in Supplementary Table II.
Comparison of the DTI metrics
As demonstrated in Figure 1, the MD values of the cingulum cingular part (CGC), cingulate gyrus (CingG), lateral fronto-orbital gyrus (LFOG), gyrus rectus (RG), as well as medial fronto-orbital gyrus in the twin group were significantly higher than the singleton group (adjusted p<0.05). The AD and RD were also increased in these regions (Figure 1), while RD showed differences in three additional ROIs, including the lingual gyrus, fusiform gyrus (Fu), and cingulum hippocampal part. The color maps in Figure 2 demonstrate the spatial distribution of regions with significant differences in MD between the groups, and the colors indicated the percentage of group difference by calculating the mean MD difference between the groups normalized to the mean MD of the singleton group for each ROI. No significant difference in FA was found between the twin and singleton groups.
Comparison of the structural connectivity
The degree, transitivity, local efficiency, global efficiency, modularity, characteristic path length, and small-worldness of the tractography-based structural network were calculated. Only the transitivity, which reflected local connectivity (39), was found to be significantly higher in the twin group than that in the singleton group (Figure 3), and the result did not change with cut-off thresholds (weakest 1-5% connections) (Supplementary Table III). Transitivity is a classical variant of the traditional clustering coefficient (40), which is thought to be less biased towards the contribution from low-degree vertices.
Comparison of the rs-fMRI results
As demonstrated in Figure 4, fALFF values in the salience network, which included bilateral insular cortex (Ins) and bilateral cingulate cortex (CingG), and several gyral regions, including the bilateral superior temporal gyrus, bilateral LFOG and bilateral RG, and left postcentral gyrus (POCG_L), were found to be significantly higher in the twin group. Moreover, the fALFF results overlapped with the MD results in several regions, including the CingG, LFOG, and RG (Figure 5). We repeated the analysis with global signal regression, and the results remained largely unchanged (Supplementary Figure 1). No group difference was found in ALFF, and no functional network parameter was found to be different in this study.
Discussion
In this study, we performed a comprehensive multi-modal MRI study to examine the structural and functional features in twin and singleton brains at term-equivalent age. Our results revealed considerable developmental differences between the two groups compared to previously reported findings, possibly due to the fact this study recruited gestational age-matched twins and singletons whereas previous studies compared preterm-born twins with term-born singletons. Although the gestational age could be included as a covariate, it is difficult to assess how well the effects of the preterm-birth and twin-birth were separated. Therefore, a direct comparison of preterm-born twins and singletons is necessary.
The volumetric differences were negligible after multiple comparison correction, which was consistent with some of the earlier studies. Knickmeyer et al. (10) compared brain volumes between twins and term-born singletons in the first month of life, and found CSF and frontal white matter volumes were greater in twins than in singletons. Ordaz et al. (11) compared brain volumes between twins and sex-matched unrelated singletons at the pediatric stage, and no significant difference was found. Another morphology study by Hulshoff et al. (9) compared monozygotic twins and dizygotic twins with their siblings in adulthood and found the difference in white matter volume diminished after correction for intracranial volume. Our finding is consistent with the latter two studies but not the first one, which is possibly related to the study population, as well as the use of different brain atlases.
For the DTI measurements, the diffusivities in several cortical gyri, especially the cingulate and front-orbital regions, demonstrated higher values in the twin group than in the singleton group. To the best of our knowledge, only one existing research studied the difference between twins and singletons using diffusion MRI. Sadeghi et al. (12) compared the longitudinal development of white matter between twins and term-born singletons from birth to 2 years of age using a non-linear mixed-effects method, and found the delay parameter of the developmental curve of AD in the anterior limb of the internal capsule and anterior corona radiata was smaller in twins compared to singletons, indicating higher AD in twins during early development. This result is consistent with ours, and their reported regions also showed differences in both FA and MD in our study before multiple comparison correction. However, our data revealed additional regions with higher diffusivities, which is again related to the study population, as well as the use of different brain atlas. Since it is known that the MD value increases with age (41-43), our results indicated a developmental delay in several cortical regions in twins compared with singletons.
For the tractography-based structural connectivity, both groups displayed small-worldness in the whole-brain networks. The transitivity, which indicates local connectivity, was found to be significantly higher in the twins than in singletons. Previous evidence indicated the clustering coefficient of the structural network (similar to transitivity) decreased with brain development from neonate to adult (44,45). However, during perinatal development, some studies found that the preterm-born neonates exhibited an age-dependent increase of clustering coefficient until about term-equivalent age (46,47). These studies indicate the network properties may change in a non-linear pattern during early development, as the brain connectomes experienced ordered strengthening of short-range connectivity followed by growth of long-range connections (48). In the present study, the PMA at scan happened to be the breakpoint of a non-linear trajectory, making it difficult to interpret the developmental difference between twins and singletons. Further longitudinal follow-up studies may be needed to understand the network difference at term-age.
In the rs-fMRI analyses, fALFF values in the salience network and several fronto-temporal regions were found to be significantly higher in the twin group, while no difference in ALFF was found. The discordance of the two parameters may be due to the artifact generated by head motion. Although sedation was used, neonatal subjects still showed more pronounced motion artifacts than adults, which may not be entirely eliminated by motion correction. The fALFF is a normalized version of ALFF, which is thought to be more robust against physiological artifacts and more sensitive to biological difference (34,49). Few studies have investigated the relationship between low-frequency fluctuation and brain development. Bray (50) studied children aged 7-18 years and found that age did not have a significant effect on global fALFF, but the fALFF in the salience network regions demonstrated an age-related decline. Although no neonatal study of fALFF was found, given the above evidence, our results indicate that preterm-born twins with higher fALFF demonstrated a more active spontaneous neuronal activity, which may be associated with delayed neurological development.
It is worth noting that we found differences both in structural and functional measurements (MD and fALFF) in the prefrontal and CingG (indicated with the green arrow in Figure 5), which are involved in several higher-order cognitive functions (51-53). Although the prefrontal and CingG related cognitive functions are immature at an early age (48,54), these regions are known to be rapidly developing during the perinatal stage (55). The agreement between the structural and functional evidence reinforces the differences between twins and singletons at term-equivalent age. These differences in brain development may be associated with the early feeding problems that are more frequent for preterm-born twins, as well as other perinatal-neonatal issues in twins. No difference in the functional network was found, which is slightly different from the result of the structural network. As the structural network is known to develop prior to the functional network (48,56,57), discordant findings between the structural and functional networks are within expectation (58-60). Moreover, in the current study, both the preterm twins and singletons showed relatively low levels of the functional network due to the immature functional activity in the neonatal brain as well as the use of sedation, so group difference may be difficult to detect.
The current work possesses several limitations. First, this study lacked term-born neonates as a healthy control group. A number of studies have shown that preterm-birth resulted in delayed brain development compared with term-born individuals in terms of DTI metrics (61,62) and brain volumes (63,64), and therefore, we can readily assume that both the preterm-born twins and singletons have altered DTI and volumetric measurements in comparison to healthy controls. Second, information on interventional operations for the preterm-born neonates was absent in this study, which may play a role in interpreting group differences. Although we have excluded those neonates with acquired brain injury on MRI and those with known perinatal diseases, and the Apgar scores at 5-mins after birth were relatively normal (9.83±0.38 and 9.74±0.53 for twins and singlets, p=0.46), this evidence cannot guarantee the healthy condition of the study subjects. Third, we did not investigate the correlation between clinical assessments and MRI in this study. In fact, we followed some of the subjects and performed the Bayley tests at 12 months of age, and no behavioral difference was found between groups. The high drop-out rate and insufficient number of subjects (6 twins and 20 singletons) in this study were not appropriate for statistical analysis. In addition, the number of twins was limited compared with singletons, and therefore, we performed a permutation test for the ANCOVA analysis to overcome imbalanced data in the present study (37). Finally, chloral hydrate was used in this study for sedation. Although sedation is considered safe and is frequently used in infants to reduce head motion in fMRI (65), a previous study has shown that sedation can induce a reduction in brain activity in infants (66). This could have had an impact on the rs-fMRI results. Future studies should consider and mitigate for this factor to improve the reliability of their results. Nevertheless, this was the first attempt to systematically evaluate both the structural and functional differences between twins and gestational-age matched singletons, and the DTI and rs-fMRI results collectively implied a developmental delay in twins at term-equivalent age.
Conclusion
In summary, the multi-modal MRI approach provided comprehensive information about the developmental differences between twins and singletons, compared with the previous single modal approach. Our results demonstrated that the preterm-born twins had higher MD in several cortical regions compared with gestational-age matched singletons at term-equivalent age. In addition, transitivity of tractography-based structural network and fALFF in the salient network were found to be higher in the twin group. These structural and functional differences collectively indicated that preterm-born twins may have delayed brain development compared with gestational age-matched singletons at term-equivalent age, which may be related to perinatal- neonatal problems. Therefore, MRI findings from twin studies on brain development should only be cautiously generalized to singletons, especially at term-equivalent age.