Diffusion MRI of the infant brain reveals unique asymmetry patterns during the first-half-year of development

The human brain demonstrates anatomical and functional lateralization/asymmetry between the left and right hemispheres, and such asymmetry is known to start from the early age of life. However, how the asymmetry changes with brain development during infancy remained unknown. In this study, we aimed to systematically investigate the spatiotemporal pattern of brain asymmetry in healthy preterm-born infants during the first-half-year of development, using high angular resolution diffusion MRI. Sixty-five healthy preterm-born infants (gestational age between 25.3-36.6 weeks) were scanned with postmenstrual age (PMA) ranging from term-equivalent age (TEA) to 6-months. At the regional level, we performed a region-of-interest-based analysis by segmenting the brain into 63 symmetrical pairs of regions, based on which the laterality index was assessed and correlated with PMA. At the voxel level, we performed a fixel-based analysis of each fiber component between the native and left-right flipped data, separately in TEA- 1 month, 1-3 months, and 3-6 months groups. The infant brains demonstrated extensive regions with structural asymmetry during their first half-of-year of life. A distinct central-peripheral asymmetry pattern was observed in mean diffusivity, namely, leftward lateralization in the neocortex and rightward asymmetry in the deep brain regions. Besides, the posterior brain demonstrated a higher lateralization index compared with the anterior brain in all metrics, which is congruent with the brain developmental pattern from caudal to rostral. Regionally, language processing regions showed a rightward asymmetry, while visuospatial processing regions exhibited leftward lateralization in fractional anisotropy, fibre density, and fibre cross-section measurements, and most white matter regions were lateralized to the left in these measurements. The laterality index of several regions (12 out 63) demonstrated significant developmental changes in mean diffusivity. At the fixel level, the fiber cross-section of inferior fronto-occipital fasciculus showed significant leftward asymmetry and the extent of asymmetry increased with PMA. In summary, the results revealed unique spatiotemporal patterns of macro- and micro-structural asymmetry in early life, which dynamically changed with age. These findings may contribute to the understanding of brain development during infancy.


Introduction
The left and right hemispheres of the human brain demonstrated anatomical and functional lateralization (also termed asymmetry) with high variability between individuals, which is associated with functional specialization in cognitive processes ( Duboc et al., 2015 ;Kong et al., 2018 ;Toga and Thompson, 2003 ). For instance, language processing is lateralized to the left cerebral hemisphere ( Duboc et al., 2015 ; the genetic effects on brain asymmetry while the environmental factors are relatively minor. Neuroimaging studies of brain asymmetry during infancy are limited so far, mainly focusing on volumetric measures, and the findings were inconsistent ( Dean et al., 2018 ;Gilmore et al., 2007 ;Lehtola et al., 2019 ). For instance, Gilmore et al. ( Gilmore et al., 2007 ) studied volumetric asymmetry in neonates during the first few weeks after birth (42.8 ± 1.6 weeks) and found the left hemisphere was larger than the right. Another two infant studies ( Dean et al., 2018 ;Lehtola et al., 2019 ) reported a larger volume of the right hemisphere, and none of them found significant age-dependence for the asymmetry indices, possibly due to a narrow age range (34.1 ± 7.7 days in ( Dean et al., 2018 ), 2-5 weeks in ( Lehtola et al., 2019 )). A few groups investigated brain asymmetry during the early period of life using diffusion tensor imaging (DTI) and found asymmetries in several white matter (WM) regions, yet no consensus has been reached ( Dean et al., 2017 ;Dubois et al., 2009 ;Liu et al., 2010 ). Dubois et al. showed a leftward asymmetry in the arcuate fasciculus and cortico-spinal tract in terms of fractional anisotropy (FA) during 1-4 months ( Dubois et al., 2009 ). Dean et al. reported a leftward asymmetry in FA of the corpus callosum, posterior limb of internal capsule, external capsule, etc., but rightward lateralization in the anterior limb of the internal capsule, stria terminalis, and the superior fronto-occipital fasciculus in infants at 1-month-old ( Dean et al., 2017 ).
Besides the heterogeneous age ranges, the methodology employed in existing studies also makes it difficult to understand the spatiotemporal development of asymmetry across the whole brain. For instance, the volumetric studies only focused on the grey matter (GM) ( Dean et al., 2018 ;Gilmore et al., 2007 ;Lehtola et al., 2019 ), while the DTI metrics were usually evaluated in the WM ( Dean et al., 2018 ;Lehtola et al., 2019 ). Moreover, DTI is a powerful technique and widely used for assessing WM integrity, but fails to resolve the within-voxel crossing fibres or provide specific information about the tissue microstructures ( Douaud et al., 2011 ;Jeurissen et al., 2013 ;Pierpaoli et al., 2001 ). By acquiring diffusion with multiple gradient orientations and b-values in the q-space, namely, high angular resolution diffusion imaging (HARDI) ( Tuch et al., 2002 ), we are able to resolve multiple microstructural components within a voxel and to characterize the microstructural properties ( Assaf et al., 2008 ;Kaden et al., 2016 ;Zhang et al., 2012 ). Fixelbased analysis (FBA) is a recently developed model-free approach to identify crossing fibres with each fibre component in a voxel called a "fixel " ( Raffelt et al., 2017 ;Tournier et al., 2004 ) and provides microstructural measurements including the fibre density (FD) and fibre cross-section (FC) that reflect the intra-axonal fibre morphology and organizational properties of each fixel ( Raffelt et al., 2015( Raffelt et al., , 2017. The first half-year of life is a critical stage of rapid brain development. The brain asymmetry pattern is already started in this period but how it changes with brain development is not fully investigated to the best of our knowledge. Two major hypotheses are brought up in this study: (1) brain structural asymmetry starts from early infancy at both macroscopic and microscopic scales and could be unveiled by diffusion MRI; (2) the asymmetry would show spatiotemporal changes with brain development in the first half-year of life. In order to test theses hypotheses, we systemically investigate the structural asymmetry of infant brains at both macroscopic and microscopic levels using HARDI data and characterize the spatiotemporal development of whole-brain asymmetry from term-equivalent age (TEA) to six months, using both the region-of-interest (ROI)-based and fixel-based methods.

Subjects
Healthy preterm-born infants were enrolled from TEA to postmenstrual age (PMA) of six months old for the MRI scan at the Children's Hospital of Zhejiang University School of Medicine. Ethical approval was obtained from the Institutional Review Board at the local hospi-tal. Written informed consent was provided by the parents or legal guardians. Exclusion criteria included (1) congenital malformation or syndrome; (2) encephalopathy caused by various factors; (3) intrauterine growth restriction; (4) acquired punctate or focal lesions, marked dilation of the cerebral ventricles on MRI (assessed by a radiologist T.L.); (5) visible artifacts on MRI; (6) psychiatric or neurological family history; and (7) illicit drug and alcohol during pregnancy.

Image acquisition
All neonates received 50 mg/kg oral or enema chloral hydrate 30 min before scanning by a nurse who was trained and certified to administer sedation. Ear protectors were used for protection. A vacuum immobilization mat was used to minimize motion. Heart and respiratory rates were continuously monitored through the scanner's physiological monitoring system. In addition, a neonatologist was always present in the scanner room and was ready for first aid. Multi-shell HARDI data were acquired on a Philips 3.0T Achieva system (maximum gradient strength of 80 mT/m and maximum slew rate of 200 mT/m/s) with an 8-channel head coil, using a single-shot echo-planar imaging sequence with phaseencoding along the anterior-posterior direction. An additional b0 image was acquired with phase-encoding along posterior-anterior direction for eddy correction. The detailed parameters were as following: diffusion time/duration = 58.5/20.6 ms, two b values of 800/1500 s/mm 2 , 32 noncolinear diffusion encoding directions for each b-value, two b0 images, repetition time = 9652 ms, echo time = 115 ms, in-plane resolution = 1.5 × 1.5 mm 2 , field of view = 180 × 180 mm 2 , 60 slices at a slice thickness of 2 mm, SENSE acceleration factor of 2, bandwidth = 1341 Hz/pixel, and echo spacing = 0.98 ms. The coil sensitivity profiles were automatically corrected via the SENSE implantation on the scanner.

DTI preprocessing
All diffusion data were preprocessed according to the Developing Human Brain Connectome Project (dHCP) pipeline ( Bastiani et al., 2019a ), including intra-subject registration using a linear image registration tool "flirt " ( Jenkinson et al., 2002 ;Jenkinson and Smith, 2001 ), followed by distortion correction and eddy current correction using "topup " and "eddy " tools in FSL ( Andersson et al., 2003 ;Andersson and Sotiropoulos, 2016 ). Motion parameters were extracted using an automated quality control tool "eddy_qc " of FSL ( Bastiani et al., 2019b ). FA and mean diffusivity (MD) maps were generated from the diffusion tensors using the weighted linear least squares method ( Basser et al., 1994 ).

Brain segmentation for ROI-based analysis
First, the mean diffusion-weighted images (DWIs) of subjects were registered to the JHU-neonate single brain DWI atlas ( Oishi et al., 2011 ) using affine transformation, followed by histogram matching between the subject images and the atlas. Then, multi-channel large deformation diffeomorphic metric mapping (LDDMM) was performed using the mean DWI, FA, and MD contrasts for non-linear transformation ( Christensen et al., 1996 ;Djamanakova et al., 2013 ;Miller et al., 1993 ). Next, the JHU-neonate single brain parcellation map, which included 63 paired ROIs, was back-transformed into the individual native space. Finally, mean values of all diffusion metrics were extracted from the ROIs ( Fig. 1 ), and an MD threshold of 2 × 10 − 3 mm 2 /s was used to exclude the cerebrospinal fluid (CSF) voxels.

FBA pipeline for asymmetry analysis
The preprocessed DWIs were left-right flipped, and FBA was performed between the native and the flipped data following MRtrix3 ( https://www.mrtrix.org/ ) processing pipeline ( Raffelt et al., 2017 ). First, response function representing single-fibre WM was obtained using multi-shell multi-tissue constrained spherical deconvolution (MSMT-CSD) method ( Jeurissen et al., 2014 ) to estimate the fibre orientation (1) For ROI-based analysis (pipeline in the red box), segmentation of the DTI data was achieved by transforming individual subjects to the JHU-neonate single brain atlas utilizing multi-channel contrasts of FA, MD, and mean DWI. Mean FA and MD in sixty-three paired ROIs were extracted for pairwise comparison between the left and right ROIs, and for correlation analysis between LI and PMA at the scan. (2) For FBA, the preprocessed DWIs were left-right flipped, and FBA was performed between the native and flipped data. Participants were categorized into three age groups of TEA-1 month, 1-3 months, and 3-6 months. For each group, a FOD template was generated, based on which the FD and FC were calculated. Statistical analysis was performed to identify the fixels with significant group differences. At the same time, the sum of FD and mean FC in each voxel were extracted for ROI-based analysis. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) distributions (FODs) on the upsampled data (1 mm isotropic resolution). As the infant brains experience rapid development in the first half of the year, we categorized these participants into three groups based on the PMA at scan: TEA to 1 month (39.9-44 weeks), 1-3 months (44-52 weeks), and 3-6 months (52-64 weeks), and an age-specific FOD template was generated for each group. Note that the FOD template for each group is symmetric since it is generated with both the native and flipped WM FODs of each individual using the "population_template " function in MRtrix to obtain an unbiased group-average template via FOD-based registration of all images to a midway space. Next, WM FODs of individual subjects were transformed to the corresponding template, followed by fixel estimation via FOD segmentation and fixel reorientation. Finally, FD and FC were calculated. For ROI-based analysis, the sum of FD and mean FC in each voxel were extracted to obtain the ROI averages.

Tractography on FOD templates
For each group, whole-brain tractography was generated from the corresponding FOD template using the second-order Integration over fibre orientation distributions method ( Tournier et al., 2010 ). Ten million streamlines were generated with seed voxels randomly selected across the whole-brain mask using the following tracking parameters: step size = 0.5 mm, maximum angle = 22.5°, min/max length = 10/250 mm, and cutoff = 0.05. Then streamlines were filtered to one million using spherical-deconvolution informed filtering of tractograms (SIFT) for reducing tractography bias ( Smith et al., 2013 ).

Verification experiment
To assess whether the asymmetry findings are biased by preterm birth, we compared the asymmetry results between the term-born and preterm-born infants utilizing the developing Human Brain Connect (dHCP) neonatal source data Release 2.0 ( Hughes et al., 2017 ). The dHCP data were acquired on a 3T Philips Achieva scanner using the following parameters: TR/TE = 3800/90 ms, in-plane resolution = 1.5 mm, thickness = 3 mm, overlap = 1.5 mm, a total of 300 volumes was collected for each subject, sampled using 4 different phase encoding directions and 3 different b-value shells (20 b0s, 64 b400s, 88 b1000s, and 128 b2600s).
28 paired preterm-and term-born infant data were selected from the dHCP data (PMA between 29.3 and 45.1 weeks) with the following criteria: (1) matched PMA at scan (38-45 weeks), gender, and head circumstance at scan (see demographic and clinical information in Supplementary Table 1); (2) radiological score under 3 (subjects with major brain lesions were excluded but minor lesions that were unlikely to have clinical significance were included); and (3) reasonable image quality. The detailed demographic information for the 28 pairs of infants was listed in Supplementary Table 1. The same preprocessing and processing pipelines in 2.3 for ROI-based analysis and FBA were performed to compare the LIs between the term-and preterm-born infants, with body-weight at birth and multiple/single birth as covariates. In addition, in order to evaluate whether dHCP results can be generalized to the in-house collected data, we selected 28 subjects from our dataset with PMA matched to the dHCP preterm-born neonates, and perform a correlation analysis of diffusion metrics/LIs between the two datasets.
Furthermore, we directly assess the effect of preterm birth on brain asymmetry by examining the relationship between gestational age (GA) at birth and the laterality index (LI) in each ROI for all diffusion metrics using linear regression, with gender, body-weight at birth, and PMA at scan as covariates.

Statistical analysis
Statistical tests for FBA were performed using MRtrix3 , and other statistics used the R-Project 4.0.2 ( https://www.r-project.org/ ). For the demographic information, categorical data were analyzed using the Chisquare test. The student's t -test was used for normally distributed data that satisfied the Shapiro-Wilk's test, and otherwise, the Mann-Whitney U test was applied. For ROI-based analysis, a pairwise t -test was performed between the left and right ROIs. LI was calculated for each paired ROI as ((left -right) / (left + right) * 100), with positive LI in FA/FD/FC and negative LI in MD representing leftward lateralization and the opposite results representing rightward lateralization. Then the analysis of covariance (ANCOVA) was performed for evaluating the relationship between LI and age, with gender, motion, body-weight at birth, and GA at birth as covariates. The P -values of the 63 paired ROIs were adjusted using the false discovery rate (FDR) method. Connectivity-based fixel enhancement and non-parametric permutation tests were used for evaluating the significance of fixel-wise asymmetry between the native and flipped images, with 5000 resamples for each fixel metric ( Raffelt et al., 2015 ). The significance level was set to 0.05 for all analyzes.

Demographic and clinical characteristics
Seventy-nine preterm-born infants within half-year-old were enrolled in this study. A total of 14 subjects (2 subjects with acquired brain lesions on MRI, 1 with hereditary disease, 1 with intracranial infection, and 10 subjects with poor imaging quality) who did not meet the inclusion criteria were excluded. Finally, 65 healthy preterm-born infants without unknown clinical abnormalities were included in this study, and the demographic and basic clinical information was provided in Table 1 . Note that the Apgar score at 5 min was 8.96 ± 2.16, indicating these preterm-born babies were relatively normal at birth.

Developmental trajectories
The tensor-based (FA/MD) and FBA-based (FD/FC) metrics demonstrated unique developmental trajectories ( Fig. 2 ). While MD gradually decreased with age in a non-linear fashion, FC increased almost linearly increased with age. FD showed a nonlinear developmental change following the typical Gompertz model (Gompertz, 1985) with sharp in-crease and then a pleatau starting approximately at PMA of 50-55 weeks ( Fig. 2 , last row). More importantly, we observed clear differences between the left and right trajectories in the majority of the brain regions, and the differences increased with PMA in the majority of the structures, such as the cingulum cingular part, superior longitudinal fasciculus, and entorhinal cortex.

Pairwise comparisons of ROI-based measurements
Generally, in the first half-year, the infant brains demonstrated extensive structural asymmetry in the majority of the brain regions (61.9% of the ROIs in MD, 60.3% in FA, 57.1% in FD, and 54.0% in FC) ( Table 2 ). A distinct central-peripheral asymmetry pattern was observed in MD, characterized by leftward lateralization in the neocortex (except for some regions near the midline) and rightward asymmetry in the central brain, including the thalamus, putamen, globus pallidus, posterior limb of internal capsule, external capsule ( Fig. 3 A). The asymmetry patterns in FA ( Fig. 3 B), FD, and FC ( Supplementary Fig. 1) shared similarities, but the FBA-based microstructural metrics showed a higher extent of lateralization than the DTI metrics overall. Furthermore, the posterior cortex, e.g., the precuneus and cuneus, showed more prominent lateralization compared with the anterior regions in all metrics ( Fig. 3 ).
Regionally, the superior temporal cortex, angular gyrus, supramarginal gyrus, which are related to language processing, demonstrated rightward lateralization with negative LIs in FA, FD, and FC, and positive LIs in MD. In contrast, the visuospatial processing related regions, including the parietal-occipital cortex, medial frontal gyrus, precentral gyrus, inferior temporal gyrus, fusiform gyrus, cingulate cortex, and superior longitudinal fasciculus, were found to be lateralized to the left side with positive LIs in FA, FD, and FC ( Fig. 4 and Supplementary Fig.  2). Developmentally, the language processing regions demonstrated increasing rightward structural lateralization in FA and FD, while the visuospatial processing regions showed relatively stable leftward structural lateralization (regression curves in Fig. 4 ).
Besides, the majority of the WM regions, including the external capsule, superior corona radiata, cingulum cingular part, cingulum hippocampal part, superior longitudinal fasciculus, corticospinal tract, pontine crossing tract, superior corona radiata, tapetum, etc., demonstrated leftward lateralization with positive LIs in FA, FD, and FC and negative LIs in MD, indicating WM in the left hemisphere may develop in advance of the left side.

Developmental changes of LI with PMA at scan
The LIs in several regions (12 out of 63 ROIs) demonstrated significant developmental changes with PMA, especially for the MD measurement ( Fig. 5 ). For example, MD in the precuneus, cingulate gyrus, posterior limb of internal capsule, and stria terminalis showed increasing lateralization towards negative LIs, while MD in the postcentral gyrus demonstrated positive LIs that further increased with PMA ( Fig. 5 A). Notably, we observed a shift of lateralization in the inferior frontal gyrus from positive to negative LIs in FA, FD, and FC measurements, with the zero-crossing occurring around 45 weeks for FA and 50 weeks for FD and FC ( Fig. 5 B-D).

FBA revealed a rightward asymmetry in FC of the inferior fronto-occipital fasciculus
At the fixel level, the difference of FC between the left and right WM was highlighted in the inferior fronto-occipital fasciculus (IFO), which demonstrated a rightward asymmetry across the three time periods ( Fig. 6 A-C). At TEA-1 month, several regions in the central brain regions with relatively low t-values that did not reach statistical significance. At 1-3 months, the high t-values regions were localized to the IFO and external capsule, and the center part of the IFO (yellow arrows) showed significant asymmetry, which became more pronounced at 3-6 months with higher t-values and larger coverage along the IFO. The FBA findings were consistent with ROI-based analysis that showed significantly higher FC in the right IFO than the left side ( Fig. 6 D). However, no statistical difference in FD was found between the left and right WM.

Effect of premature birth on brain asymmetry
The dHCP data revealed no difference in asymmetry between the healthy preterm-born and term-born infants at TEA using either the ROIbased analysis ( Supplementary Fig. 3-4) or FBA, indicating the findings from healthy preterm-born infants in our study was representative of normal brain development. Also, both preterm-born and term-born infants demonstrated a rightward asymmetry in regions relating to language processing, and leftward asymmetry in regions relating to visuospatial processing, which was similar to our findings. The correlation analysis of diffusion metrics demonstrated a high similarity between the in-house collected and dHCP preterm-born infants ( Supplementary Fig.  5). The correlation coefficients of FA, MD, LI of FA, LI of MD were 0.986, 0.960, 0.650, and 0.725, respectively, suggesting the similarity between term-and preterm-born infants in dHCP was applicable to our data.
In addition, no significant linear correlation between the LIs and GA at birth was found in any of the ROIs for any of the diffusion metrics with our dataset after multiple-correction, another indication that pretermbirth (or GA) may not play a critical role in lateralization findings in our study.

Discussion
This study, for the first time, investigated the spatiotemporal pattern of brain asymmetry in infants during early brain development with comprehensive ROI-based and fixel-based analysis. Results revealed an interesting central-to-peripheral asymmetry pattern with rightward lateralization in the deep brain and leftward lateralization in the cortex based on MD measurements. We also found that (1) the posterior brain demonstrated higher asymmetry than the anterior brain, indicating a more established function specialization in the posterior brain which is in accordance with the caudal-to-rostral trend of the brain development ( Volpe, 1995 ); (2) the rightward asymmetry in language processing regions and leftward asymmetry in visuospatial processing regions, and similar findings were repeated in both term-and preterm-born neonates using the dHCP data; (3) rightward lateralization in FC of the IFO based on both FBA and ROI-based analysis. Moreover, brain asymmetry actively changes with development in terms of the degree (LI), direction (leftward versus rightward), and spatial distribution. These findings may elucidate the process of how structural asymmetry originated in the brain during the early period of life and potentially deciphers the genetic effects on brain asymmetry apart from later environmental effects.

Central-peripheral asymmetry pattern and regional asymmetry
Brain asymmetry is known to manifest very early in life, starting from the fetal stage ( Corballis, 2013 ;Hering-Hanit et al., 2001 ), and become more pronounced after birth ( Dean et al., 2017 ;Dubois et al., 2009 ;Lehtola et al., 2019 ). Our results revealed extensive structural asymmetry in the infant brain during the first half-year of life. Specifically, we found a central-peripheral asymmetry pattern in MD, characterized by a leftward asymmetry in the neocortex but a rightward asymmetry in the deep brain WM and GM. A previous animal study reported different FA patterns between neocortex and deep brain structures in the neonatal ferret, and contributed the difference to more elongated cell bodies and less differentiated axonal/dendritic processes in the early-developing neocortex, compared with neurons in the deep brain ( Kroenke et al., 2009 ). In addition, a histological study of adult rats revealed the asymmetry of the cortex is related to the different number of neurons between the hemispheres ( Galaburda et al., 1986 ).

Asymmetry of the language and visuospatial systems
Here we focused on two functional systems that are extensively studied in asymmetry research. The language processing regions showed a rightward structural asymmetry, although individual components in the system showed laterality, e.g., the SLF was lateralized to the left. These results were repeated in both preterm-and term-born neonates from dHCP data. The language was found mostly dominant to the left hemisphere based on functional MRI in both children and adults ( Gotts et al., 2013 ;Holland et al., 2007 ;Szaflarski et al., 2006 ;Vikingstad et al., 2000 ), which is partially consistent with our finding in SLF, although structural and functional asymmetries do not necessarily correlate with each other. Dubois et al. (2009) studied infants from 1 to 4 months of age using DTI, and found a leftward asymmetry in arcuate fasciculus, which partially agrees with our result of leftward asymmetry in SLF that included accurate fasciculus.
The visuospatial processing regions demonstrated leftward lateralization in terms of FA, FD, and FC. Barrick et al. (2007) studied the morphology of white matter pathways in 30 right-handed healthy individuals (20-39 years), and found a rightward asymmetry in pathways relating to auditory spatial attention, which is consistent with another adult study using a similar method ( Powell et al., 2006 ). Caeyenberghs and Leemans (2014) investigated tractography-based structural networks in 346 healthy adults, and reported a higher efficiency in visuospatial related areas in the right hemisphere. Zhou et al. (2013) studied cortical thickness in 274 healthy participants aged 5-59 years old, and found the left medial occipital lobe that involves visual function, were more developmental during pre-adolescents. All these prior studies in adults are supportive of our findings in the visuospatial system. Yet, direct evidence in the infant brains is rare, and future studies in this area are necessary to validate the current findings.

Age-dependent change of LI in several regions
The major advantage of the in-house collected data is in its wider age range (TEA-6 months) compared with the dHCP data (PMA ≤ 45 weeks). The wide age range of our in-house data enabled characterization of the rapid development of infant brains in the first half-year of life when several milestones of neurological events take place. Brain asymmetry in this age range was not reported before. In fact, some of the previous findings have reported that the developmental changes of asymmetry were not significant within the first 1-2 months of life ( Dean et al., 2018 ;Lehtola et al., 2019 ).
With our data, LIs of several GM and WM regions were found to change with age. Some regions demonstrated an increasing trend   6. Differences in FC between the left and right brain hemisphere using FBA method. (A-C) FBA results of the FC differences between left and right hemispheres in the three age groups of TEA-1 month, 1-3 months, and 3-6 months, presented in axial, coronal, and sagittal views. t -values of the statistical analysis were overlaid on the FOD template in regions with adjusted P -value < 0.2. (D) ROI-based analysis revealed a similar rightward asymmetry of FC in IFO. The color bar represents the t values. * : adjusted P -value < 0.05, * * : adjusted P -value < 0.01, * * : adjusted P -value < 0.001. (For interpretation of the references to color in this figure, the reader is referred to the web version of this article.) of asymmetry, for instance, the LIs of the precuneus and postcentral gyrus, which are associated with visual and sensory functions, demonstrated a noticeable increase of asymmetry with age. Some other regions exhibited a shift of lateralization direction with age, for example, the inferior frontal gyrus (IFG) shifted from leftward to rightward asymmetry in FA, FD, and FC measurements. The IFG, which is known to be related to language function, was found to be lateralized to the left in adult brain studies ( Kovalev et al., 2003 ;Lee et al., 2014 ;Plessen et al., 2014 ). However, a meta-analysis found that different sub-divisions of the IFG demonstrated opposite asymmetries ( Kong et al., 2018 ), which may explain the shift of lateralization in our finding.

Microstructural asymmetry in the IFO
The present research examined the microstructural asymmetry of WM in infants in a voxel-wise manner that was not reported before. The main findings revealed a rightward asymmetry in FC of the IFO across age groups, which was confirmed by ROI-based analysis. The IFO is an important WM bundle connecting the prefrontal and parieto-occipital lobe, and plays important role in auditory comprehension, attention, and affective behavior ( Altieri et al., 2019 ;Zhang et al., 2018 ). A previous study in children between 9,10 years old found that a rightward dominance in FA of the IFO ( Banfi et al., 2019 ) that is congruent with our results. Another group studied the microstructural asymmetry in adults using FBA ( Arun et al., 2020 ), and reported similar findings about IFO with ours.

Effects of preterm-birth on asymmetry
Several studies have compared the asymmetries of preterm-born and term-born individuals, but the findings are not consistent. Lancefield et al. (2006) studied the volumetric asymmetry in a group of 14-year-old individuals born very preterm and age-matched full-term controls, and reported no significant between-group differences. Two other studies examined the functional asymmetry in preterm-born and term-born neonates ( Kwon et al., 2015 ) and adults , and found the preterm-born group demonstrated significantly less lateralization compared with the term-born group at both neonatal and adult stages. The inconsistency may arise from the study population and data processing procedures. Therefore, we performed a verification experiment utilizing the dHCP release-2 data, and confirmed no differences in brain asymmetry between preterm-born and term-born infants using our data processing pipeline. Furthermore, no significant linear correlation between the LIs and GA was found. Moreover, subjects were included in this study based on a rigorous clinical criterion, and participants with visible lesions on the MRI were excluded. Although premature delivery may lead to subtle WM damage that was not detectable, the damage is more likely to be diffusive rather than asymmetric ( Back and Miller, 2014 ;Horsch et al., 2007 ;Inder et al., 2003 ). Therefore, the findings in this healthy preterm-born infant study could reflect the structural asymmetry during normal development and the findings may be generalized to term-born infants. Nevertheless, future studies in healthy term-born neonates are necessary. Also, despite the structural similarity, functional asymmetry may still be different between term-and pretermborn neonates.

Limitations
Besides the aforementioned caveat of the lack of term-born infants, the present study also has several other limitations. In our processing pipeline, we did not consider the motion-related B0 inhomogeneity variation in the current study, which could be potentially corrected ( Andersson et al., 2018 ). Also, our imaging protocol only acquired reversed phase-encoding for the b0 image but not for the DWIs considering the relatively scan time and stability of sedation in neonatal patients. On the other hand, the amount of motion estimated from our data was relatively small (the median absolute motion: 0.62 mm, std: 0.29 min: 0.2 max: 1.94 mm), and less than the dHCP data (the median absolute motion: 1.52 mm, std: 0.62 min: 0.52 max: 6.31 mm) ( Bastiani et al., 2019a ), which is unlikely to induce substantial changes in B0 and cause intractable distortion. In addition, the handedness information was absent as in other infant studies. In a previous study with a large number of adult subjects, handedness was not found to be significantly associated with asymmetry ( Kong et al., 2018 ). Last but not least, enema chloral hydrate was used in this study for sedation according to the routine scan protocol at our local hospital, which is considered to be safe ( Finnemore et al., 2014 ). Although a previous study found that sedation can induce a reduction in brain activity in infants, which may be an important confounder in functional MRI ( Williams et al., 2015 ), there is no evidence that chloral hydrate will affect the structural MRI.

Conclusions
The present study investigated microstructural asymmetry in healthy preterm-born infants using HARDI data. We found the majority of brain regions showed significant asymmetry during the first half-year of life and the asymmetry dynamically changed with brain development. The MD-based laterality index showed a central-peripheral asymmetry pattern with the most prominent lateralization in the posterior cortex and a rightward asymmetry in language processing regions and leftward asymmetry in visuospatial processing regions. The IFO demonstrated a rightward asymmetry using fixel-based microstructural analysis. These spatiotemporal changes may advance our understanding of brain asymmetry during early development, and the asymmetry patterns may provide useful markers for the evaluation of developmental disorders.