Trajectory of SLF in the human neonatal brain
In this study, we used the SS3T-CSD algorithm proposed by Dhollander et al. to calculate the WM-FOD images meeting the tracking requirements and then successfully divided SLF into SLF I, SLF II and SLF III in the neonatal brain. This is different from the previous results of many newborn DTI studies. Huang et al. used DTI technique to conduct deterministic fiber tracking of the fetal brain in the second trimester, third trimester and neonatal period respectively. Their results indicated that SLF could not be tracked in the second trimester and only part of SLF (mainly refer to AF) could be detectable in the third trimester as well as neonatal period. Therefore, they speculated that SLF might emerge in the third trimester (Huang et al. 2006; Ouyang et al. 2015,2019). In fact, such DTI fiber tracking results are likely to be incomplete. From the perspective of the developmental characteristics of SLF, as one of advanced association fibers, it matures later than other fiber tracts and has a low degree of myelination in the fetal and neonatal periods (Kulikova et al. 2015). In addition, there are corpus callosum, corticospinal tracts and multiple fiber tracts in the running regions of SLF crossing with it. While DTI is limited by the model algorithm and can simply estimates the only one major fiber distribution direction in a single voxel (Jeurissen et al. 2019). All these factors will greatly interfere with the DTI tracking results of SLF. With the development of MR sequences, more advanced diffusion MRI techniques, such as high angular resolution diffusion imaging (HARDI) and diffusion spectrum imaging (DSI) based on probability density function, have been widely applied and popularized (Tuch et al. 2002; Wedeen et al. 2005,2008). They effectively make up for the deficiencies of DTI algorithm and present more abundant and real fiber bundle direction as well as connection information, so as to fully display the fiber bundles with more complex structures such as crossing and bifurcation. However, these advanced MRI methods still not be widely applied in the neonates due to the limitation of scanning time. Recently, Hutter et al. proposed a highly optimized HARDI MRI protocol that can control the scanning time within the tolerated range of the newborns (20 minutes) (Hutter et al. 2018). This protocol helped dHCP to collect enough high-quality neonatal diffusion data. Besides, SS3T-CSD was proposed and shown to be performed well in the dHCP data (Dhollander et al. 2019). The most prominent advantage of SS3T-CSD is that it can model 3 tissue compartments well using single-shell data and thus there is no need to segment newborn brain tissue into white matter, gray matter, and cerebrospinal fluid, which avoids the segmentation difficulties of neonatal brain tissue.
Although SS3T-CSD algorithm helped us obtain the optimized WM-FOD image and high-quality whole-brain fiber tracking map, we still need to complete the delineation of ROI in order to get the final SLF segmentation results. Some researchers performed DSI-based tractography to get the adult SLF and its three branches selecting the cortical ROIs in parietal and frontal regions (Wang et al. 2016; Nakajima et al. 2019). In view of this, we also used parietal cortical ROIs to track SLF in neonatal brain, but we found that fibers from parietal lobe to the frontal lobe were interrupted, and we could not get the three branches of SLF in the prefrontal areas. Similarly, the fibers in posterior parietal regions could not be tracked using the cortical ROIs of the frontal lobe. After have ruled out many factors such as unreasonable tracking parameters (step size, angle, FOD amplitude) that may lead to fiber interruption, we found that there were many U-shaped fibers (short association fibers) in the SLF branches during the neonatal period and fiber tracking was prone to be interrupted in connection areas of adjacent U-shaped fibers. Although it has been reported that adult SLF was mainly composed of short association fibers connecting neighboring cortical regions (Catani and Thiebaut de Schotten 2012; Yagmurlu et al. 2016), we speculated that the main reason for the two different tracking results between adults and neonates may be related to the immature fiber structure of SLF in neonatal period. Axonal pruning, myelination and other maturation processes which have finished in human adult period may be conducive to fiber tracking results. In order to reduce or even avoid interruptions in the tracking process, we adopted the whole-brain fiber tracking method and got the three branches of SLF by manually plotting multiple ROIs in the white matter regions along the fibers.
From the segmentation results of SLF, we found that the cortical regions connected by SLF three branches in the human neonatal brain were similar to those of adults, which to some extent verified the accuracy of our segmentation. Actually, there are still some controversies about segmentation of SLF three branches at present. For example, several studies have questioned the existence of SLF I (De Benedictis et al. 2016; Wang et al. 2016; Nakajima et al. 2019; Schurr et al. 2020). Some researchers believe that SLF I has many similarities with the cingulum in anatomical structures and functions, so SLF I belongs to the cingulum fiber system (Wang et al. 2016; Nakajima et al. 2019). In contrast, Yagmurlu et al. demonstrated that SLF I was an independent fiber bundle using both DSI fiber tracking and microanatomical dissection technique (Yagmurlu et al. 2016). Besides, Komaitis et al. successfully separated SLF I and cingulum on human adult specimens, likewise, proving that they were independent fiber systems (Komaitis et al. 2019). Our results also showed that SLF I was located above the cingulum and the two fiber tracts were clearly separated by the corpus callosum on the three-dimensional fiber tracking map, although SLF I and the cingulum were close to each other in the cross-section planes. In addition, it is occasionally difficult to separate SLF II and SLF III on the axial planes (Makris et al. 2005). Combined with a lot of relevant literatures, we summarized our own segmentation principles. That is in the sagittal planes where SLF II and SLF III exist simultaneously, the fiber bundles running with a more dorsal direction were regarded as SLF II, while the remaining fibers with a more ventral direction were regarded as SLF III (Wang et al. 2016).
Quantitative analysis of SLF subcomponents
After obtaining the fiber tracking results of SLF three branches in 40 neonates, we measured the DTI and NODDI parameters values of each branch and conducted quantitative analysis on their maturation degree.
Our statistical results showed that there were no side and sex differences of parameters values in the SLF three branches. Earlier DTI studies on development of SLF also got the same results (Liu et al. 2010, 2011). While many studies have reported that the SLF of human during the adult period shows obvious lateralization and asymmetry (Makris et al. 2005; Thiebaut de Schotten et al. 2011b; Budisavljevic et al. 2017; Nakajima et al. 2019). Therefore, we predicted that the side and gender differences of SLF were not obvious or present in the newborn period. With the development and maturity of fiber function, these differences may gradually show up during the later period.
We also performed linear regression analysis of DTI and NODDI parameters values in relation to gestational age. Among these parameters, FA is one of the most commonly used metrics to evaluate the maturity of fiber tracts (Qiu et al. 2015; Feng et al. 2019), which reflects the degree of fractional anisotropy of fiber bundles (Pierpaoli and Basser 1996). Our results showed that the FA values of SLF I, SLF II and SLF III all increased linearly with the increase of gestational age. This indicated that the fractional anisotropy of the SLF three branches enhanced in neonatal period. However, from the point of view of biophysical properties, the increase of fractional anisotropy is closely related to the changes of many white matter microstructure characteristics, such as axonal density, axonal diameter, cell membrane permeability, myelination degree, axonal orientation distribution, etc. (Jones et al. 2013). Any variations of these factors can cause the change of FA value. In order to further explain the reason for the increase of FA value of fiber tracts, we also observed the changes of NODDI parameters values. The NDI value represents the neurite density, while the ODI value represents the neurite orientation dispersion, especially the degree of fiber coherence (Dean et al. 2017; Genc et al. 2017). Therefore, the increase of NDI value or the decrease of ODI value can lead to the increase of FA value (Chang et al. 2015). Our results showed that with the increase of gestational age, the NDI values of SLF I, SLF II and SLF III all increased linearly. While the ODI values of SLF II decreased linearly, and the ODI values of SLF I and SLF III changed nonlinearly. In view of this, we speculated that the increase in FA values of SLF I and SLF III during the neonatal period was mainly due to the increase of axonal density, with little influence by the change of axonal orientation dispersion. The increase of FA value of SLF II was influenced both by the increase of axonal density and the decrease of axonal orientation dispersion in this period. Our further analysis of NODDI parameters values revealed two key factors that may lead to the changes in the FA values of SLF three branches, thus improving the biological specificity of the FA value. Besides, other DTI parameters (RD, AD, MD) also changed differently in the three fiber branches: with the increase of gestational age, the MD, RD, and AD values of SLF I as well as SLF III decreased linearly. The MD and RD values of SLF II also decreased linearly, but the AD values of it changed nonlinearly. Among these parameters, in the biological sense, AD value can reflect the degree of axonal arrangement, while RD value represents the degree of myelination and the growth of oligodendrocyte (Song et al. 2005; Dubois et al. 2008). The results from our studies clearly showed that, with the increase of gestational age, the degree of axonal packing and the number of oligodendrocytes around the axons of SLF I and SLF III increased during this period. This led to the decrease of longitudinal and transverse diffusivity of SLF I and SLF III and finally caused the reduce of mean diffusivity. However, SLF II was different. The decrease of mean diffusivity of it was mainly caused by the decrease of transverse diffusivity, which means that the number of oligodendrocytes around the axon increased.
Moreover, the parameters values of SLF three branches have significant differences by statistical analysis. Our results showed that, among the three fiber branches, SLF III had the highest FA value, followed by SLF I, and SLF II had the lowest FA value. On the contrary, SLF II had the highest ODI value and SLF III had the lowest ODI value. In addition, the differences of NDI, MD and RD values among the three fiber branches were statistically significant. However, further multiple comparisons showed that only the parameters values differences between SLF II and SLF III were statistically significant, which showed that the NDI values of SLF III were greater than SLF II, and the MD values as well as RD values of SLF II were both higher than SLF III. According to the differences of parameters values of these fiber branches and the biological significance represented by them, we speculated that the three branches of SLF had different maturity degree. One of the hallmarks of white matter maturation after birth is retraction and myelination of the axonal pathways (LaMantia and Rakic 1990; Kunz et al. 2014), which means the decrease of the axonal orientation distribution and increase in the number of oligodendrocytes, corresponding to the increase of NDI values and decrease of ODI and RD values. The increase of NDI and the decrease of ODI led to the increase of FA, and the decrease of RD also led to the decrease of MD. Therefore, we speculated that SLF III with the highest FA values and the lowest ODI values had the highest maturity degree, while SLF II with the lowest FA value and the highest ODI value had the lowest maturity degree. The study of Horgos and colleagues also confirmed our findings. They used anatomical dissection technique to investigate the development of white matter tracts in human fetal brain aged between 13 and 35 gestational weeks. Their results showed that the periinsular portion of SLF, corresponding to our SLF III segment and AF, can be identified by dissection at 14 weeks. Then the newer fibers always emerge in more superficial positions with the gestational age increasing and formed a more superficial component of SLF (Horgos et al. 2020), corresponding to our SLF II segment. The sequence of development in periinsular area from ventral to the dorsal direction supported our quantitative analysis results, that is, ventral SLF III is more mature than dorsal SLF II in human neonatal period.
The general sequence of development is from the simple to the complex as well as from the lower to the higher, and the differences in the maturity status of these three fiber branches may be related to their different levels of brain functions. Parlatini et al. found that the functions of the human frontal and parietal regions can be divided into dorsal spatial/motor and ventral non-spatial/motor networks, corresponding to SLF I and SLF III functional networks respectively. Whereas, SLF II is associated with a network of multimodal region at the intersection between the dorsal and ventral networks. There may be many neurons with very flexible response characteristics in the regions connected by SLF II, reflecting the function of human conscious processing (Parlatini et al. 2017). In addition, some studies suggested that SLF I mainly connect the core areas of the dorsal attention network (DAN), and the SLF III mainly connect the areas of ventral attention network (VAN). While the middle SLF II, however, is reported to directly connect the DAN and the VAN, which may provide critical anatomical communication for the two attention systems (Thiebaut de Schotten et al. 2011a; Sani et al. 2019; Suo et al. 2021). Therefore, the later development tracts such as the SLF II may contribute more to higher-level brain functions especially in mediating between the SLF I and SLF III.
Limitations and future directions
We segmented the neonatal SLF into three branches and assessed quantitatively the maturation status of each branch using diffusion MRI technique. None of this has been deeply explored before. However, one of the limitations in our study is that the definition of each individual cortical ROIs was based on the registered newborn AAL atlases, whose precision and accuracy need to be further improved. In addition, although diffusion MRI tractography is currently the only non-invasive detection method of the white matter structure in vivo, our results still need to be histologically verified for the widely acknowledged limitations of tractography. And we will supplement and confirm our studies in the future through the fiber dissection technique of newborn specimens.