Obstetric factors (i.e., multiple pregnancy status, intrauterine growth restriction –IUGR, maternal antenatal steroids and magnesium therapy, delivery method) as well as infant characteristics at birth (i.e., Apgar scores at 1 and 5 min, birth weight, length, and head circumference) were extracted from the dHCP records.

Specific postnatal risk factors previously recognized to be related with neonatal brain abnormalities, including diffuse and regional WM microstructural alterations (Pogribna et al., 2013; Brouwer et al., 2017; Barnett et al., 2018; Parikh et al., 2021), were also considered. These included NICU variables (i.e., total duration of ventilatory support and oxygen therapy, and parenteral nutrition) which were binarized using thresholds established in previous studies (need of mechanical ventilation beyond 7 days, and of parenteral nutrition longer than 21 days) (Brouwer et al., 2017). Sepsis was considered as any situation where an infant received antibiotics, as there was not enough information to retain only confirmed episodes of postnatal sepsis. Additionally, we derived a neonatal morbidities binary factor to summarize the presence of at least one of the following 4 morbidities associated with prematurity (or the absence of all 4): chronic lung disease, necrotizing enterocolitis (NEC), retinopathy of prematurity (ROP) and abnormal cranial ultrasonography (cUS). Of note, detailed neonatal medical records were available only for infants admitted to the neonatal intensive care unit (NICU) after birth: 51 PT (86%) and 2 FT (3%, admitted for sepsis treatment, without further complications).

Comparisons between PT and FT groups in terms of the described variables and factors were performed with suitable tests (Wilcoxon rank sum test for ordinal and continuous variables; Fisher’s exact test for binary factors; Pearson’s Chi-squared test for non-binary nominal factors) in R (version 4.0.5, 2021.03.31).

Family socio-economic status (SES) was measured using the index of multiple deprivation (IMD) which is a UK geographically defined composite social risk score comprising data on income, employment, health, education, living environment, and crime calculated from the mother’s home address at the time of birth.

Neurodevelopmental outcome was assessed at St Thomas’ Hospital, London by two experienced assessors using the Bayley Scales of Infant and Toddler Development, Third Edition –BSID-III (Bayley, 2006). We only considered assessments performed at around 18 mCA (between 17 and 21 m), which was available for 44 (75%) PT infants and 53 (90%) FT infants (median age: 18.3 m). Five distinct developmental categories (cognition, receptive and expressive language, and fine and gross motor function) were assessed yielding age-standardized respective scaled scores (mean 10, SD 3), with higher values indicating better infant development and scores lower than 7 indicating developmental delay in that domain.

Comparisons between PT and the FT groups were performed with t-tests corrected for multiple comparisons using Benjamini–Hochberg false discovery rate (FDR) correction across scores. The effect of sub-groups (PT_EV, PT_ML, FT_EVCt, and FT_MLCt) on neurodevelopmental outcomes was assessed using one-way ANOVA.

Of note, the 5 BSID-III scaled scores can be summarized into the widely used 3 composite cognitive, language, and motor scores (mean 100, SD 15). The results of the entire analyses performed using them can be found in the Supplementary materials: BSID-III composite score results section.

The DTI model was fitted to the diffusion data using a single shell (b = 1,000 s/mm²) and calculated with FSL’s DTIFIT. The choice of using only a single b-value was made because the utility of including more diffusion directions may be outweighed by the non-Gaussian contribution of high b-value acquisitions (Pines et al., 2020). DTI maps were computed for four metrics: fractional anisotropy (FA) and Mean Diffusivity (MD) which are a composite of axial diffusivity (AD) and radial diffusivity (RD).

Additionally, multi-shell diffusion data were used to derive the NDI and ODI maps from the NODDI model (Zhang et al., 2012) using the CUDA 9.1 Diffusion Modelling Toolbox (cuDIMOT) NODDI Watson model implementation for GPUs (Hernandez-Fernandez et al., 2019). We used the MCMC optimization algorithm and default settings to fit the NODDI model to our infant data. The NODDI-derived maps were then post-processed to reduce the observed noise. Briefly, we used ODI maps to detect possible errors using an alpha-trimming strategy. The voxels presenting values outside the threshold range (fixed upper value of 0.95 and the lower limit being the first groove of the histogram of values) were either (i) normalized by the immediate surrounding values (i.e., the mean of the voxel’s immediate environment after the removal of extreme values), or (ii) set to 0, if no voxels in the “normal” range were found in their environment. The same erroneous voxels were also corrected in NDI maps in the same fashion.

Pre-processed structural data was used to parcellate 13 ROIs (6 in each hemisphere and 1 bilateral, Figure 1A) relevant for the developing SM network, focusing on the primary core of cortical and deep GM structures. Regarding cortical regions, we considered primary sensory and motor cortices which are essential for processing peripheral somatosensory inputs and for initiating and controlling motor behaviors, through cortico-subcortical loops including thalamus, BG and brainstem. We then disregarded non-primary cortical areas, which mature later during typical development. Three cortical ROIs were thus defined on the cortical surface of each hemisphere using the M-CRIB-S surface-based parcellation tool optimized for the term-born neonates (Adamson et al., 2020) whose labeling scheme replicates the Desikan-Killiany-Tourville (DKT) atlas (Klein and Tourville, 2012): the postcentral gyrus as the anatomical proxy of the lateral portion of the primary somatosensory cortex (hereafter referred to as S1 for the sake of simplification), the precentral gyrus as the lateral portion of the primary motor cortex (referred to as M1), and the paracentral lobule (referred to as ParaC) corresponding to the medial surface of the hemisphere in the continuation of the precentral and postcentral gyri, including the medial portions of the primary SM cortices. The central sulcus was required as a landmark to delineate these pre- and post-central regions, explaining the choice of a surfacic parcellation tool. Individual surface ROIs were then projected to the cortical ribbon defined in the anatomical volumes, and further dilated by one voxel into the WM to ease the tractography process.

Regarding sub-cortical structures, the most relevant to be studied at early developmental stages are the main input/output and relay structures, implicating in particular the brainstem, the thalamus, the dorsal striatum (composed of the putamen and caudate nucleus) that can be considered as the main BG input structure for SM projections, and the internal segment of the globus pallidus (GPi), one of the major output structures of the BG (Leisman et al., 2014). Identifying these specific GM structures on MR images is quite challenging in infants due to the inter-subject variability and rapid changes in morphological characteristics and sizes during the perinatal period. In addition, the precise segmentation of structures of interest also depends on the possibilities offered by the tool validated for neonatal population. The subcortical ROIs were thus defined using a volumetric GM parcellation based on Draw-EM algorithm segmentation (Makropoulos et al., 2014) provided within the dHCP data release, namely medial brainstem (Bstem) and for each hemisphere: thalamus (Thal, fusing high and low intensity regions), caudate nucleus (Caud, part of the striatum) and lenticular nucleus (Lenti, containing the putamen as well as the GPi).

These cortical ROIs (bilateral M1, S1, and ParaC region) and subcortical ROIs (brainstem and bilateral thalamus, caudate and lenticular nuclei), used as seeds for the tractography, were aligned to the diffusion space with FSL 6.0’s FLIRT.

Individual dissections of SM connections, which to our knowledge have never been achieved in neonates and infants until now, were performed using an automated tractography-based approach benefitting from multi-shell MRI data. For each subject, probabilistic tractography estimating multiple diffusion orientations within a voxel (Behrens et al., 2007) was used to reconstruct connections between the selected ROI pairs (designated as tracts thereafter). Briefly, for each subject, we first modeled crossing fibers within each voxel of the multi-shell diffusion data using a GPU accelerated version of FSL’s Bayesian Estimation of Diffusion Parameters Obtained using Sampling Techniques modelling Crossing Fibres (BEDPOSTX), with default settings apart from the deconvolution model with zeppelins (Hernández et al., 2013). Then, the pre-selected ROIs were used as seed masks to derive region-to-region structural connections using the GPU implementation of the Probabilistic Tractography with crossing fibers (ProbTrackX) available with FSL 6.0 (Hernandez-Fernandez et al., 2019), and the default (one-way) setting with a loop check. The resulting output describe the density of WM connections between the ROI pair.

To improve the tractography results, and to reduce the incidence of erroneous streamlines, we employed exclusion masks. These exclusion masks were based on a mask of CSF created by thresholding the MD maps (voxels with MD > 2.10^–3mm²/s were considered as CSF) and corrected by removing voxels with FA > 0.25 (which might correspond to WM voxels in the corpus callosum but close to the ventricles with CSF partial volume effects). The exclusion masks were further adapted to exclude all other brain structures apart from the considered ROIs pair. Additionally, where the pair of ROIs were ipsilateral, i.e., in a single hemisphere, the entire contralateral hemisphere was also excluded. No supplementary constraints were included in the tractography runs.

Reconstructed tracts were then thresholded at 5% of the maximum fiber density of the evaluated tract. This was not performed for cortico-cortical inter-hemispheric tracts, whose reconstructions were used in their original state due to low streamline numbers.

The final list of SM tracts of interest (corresponding to homotopic inter-hemispheric tracts, short-range S1–M1 intra-hemispheric tracts, and long-range intra-hemispheric cortico-subcortical tracts) is described in Figure 1B We visually validated the accuracy of the tracts reconstructions for several subjects and observed expected topographies (e.g., the S1 and M1 projections toward the ventral anterior and lateral portions of the thalami). Note that the (inter- and intra-hemispheric) cortico-cortical tracts involving paracentral regions could not be evaluated due to frequent and variable tractography errors identified upon visual examination. Also, connections between subcortical structures were not studied because of their proximity which could alter the tractography performance.

DTI and NODDI-derived metrics (FA, MD, AD, RD, NDI, ODI) were extracted from each individual tract by calculating the weighted average value (metric X¯) using the following equation:

where i denotes the tract voxels, d_i is the fiber density at voxel i of a tract, and X_i is value of the metric at voxel i (Hua et al., 2008). This weighted approach gave more weight to the central portion (with higher fiber density) compared to the tract periphery, rendered the measures independent on the number of streamlines assessed by the tractography algorithm, and limited the effect of potential artifacts related to tractography reconstruction.

To further evaluate the differential effect of prematurity on specific tracts, we subsequently compared distances for each pair of tracts through paired t-tests in each PT subgroup independently (Supplementary Figure 3). Many more significant tract-by-tract differences were observed in the PT_EV than in the PT_ML group (69 vs 26/105 for set 1; 70 vs 22/105 for set 2; 61 vs 23/105 for set 3), but the trends were rather consistent between the two PT subgroups.

Focusing on the PT_EV subgroup, the significant differences between tracts assessed by the paired t-tests allowed us to propose an ordering of the tracts based on the relative effects of prematurity on microstructural characteristics (Figure 4B). For sets 1 and 2 (DTI sets), the orderings were highly similar, with, somewhat schematically, the following tracts showing the lowest to highest distances: (1) M1-Bstem; (2) ParaC-Bstem, S1-Bstem, M1-Thal, ParaC-Thal, M1-Caud, ParaC-Caud; (3) M1-Lenti, ParaC-Lenti, S1-Thal, S1-Caud; (4) S1-Lenti, S1_L-S1_R, M1_L-M1_R; (5) S1–M1. For set 3 (NODDI set), the ordering showed schematically, from the lowest to highest distances of tracts: (1) M1-Bstem, ParaC-Bstem; (2) S1-Bstem, M1-Thal, ParaC-Thal, M1-Caud, ParaC-Caud, S1-Caud; (3) S1-Thal, M1-Lenti, ParaC-Lenti, S1_L-S1_R, M1_L-M1_R; (4) S1-Lenti, S1–M1 (Figure 4B).

Despite a few differences in the ordering of a couple of tracts, the results were quite consistent across the three sets and revealed a differential impact of prematurity on the SM tracts microstructure. Overall, the tract ordering based on maturational distances highlighted a coherent caudo-rostral and central-to-periphery pattern, with: the cortico-Brainstem tracts presenting the lowest distances and thus the least impact of prematurity; the cortico-Thalamic and cortico-Caudate tracts showing “intermediate” distances; the cortico-Lenticular tracts appearing with the highest distances among the cortico-subcortical tracts; and the cortico-cortical tracts revealing the highest prematurity impact. Inter-hemispheric tracts (S1_L-S1_R and M1_L-M1_R) showed lower distances than the intra-hemispheric tracts (S1–M1). S1 tracts generally had higher distances than the M1 and ParaC tracts, both presenting similar profiles.

This approach of tract ordering based on the maturational distance related to prematurity was not considered for the PT_ML subgroup as tract pairwise comparisons were less systematically significant than in the PT_EV group and the ordering was more difficult to synthesize. For this subgroup, the distances of all tracts were more homogeneous (Supplementary Figure 3), which may be associated with a lesser effect of prematurity on the microstructural profiles of the SM tracts.

Finally, we assessed whether maturational distances related to prematurity at TEA might be related to outcome (BSID-III scaled scores) at 18mCA, considering each PT subgroup independently. Pearson correlations showed significant results only in PT_EV group (N = 24), for Set 3 (NODDI) (Supplementary Table 4) and for specific tracts: M1-Bstem and ParaC-Bstem distances were both negatively correlated with Cognitive scaled score and Fine motor score, while M1-Lenti, ParaC-Lenti and S1–M1 tracts were also negatively correlated with Fine motor score (the lower the maturational distance, the higher the score and thus the better the outcome) (Figure 5). Interestingly, these five tracts showed different levels of distances over the PT_EV group, with M1- and ParaC-Bstem having a distance closest to 0, whereas S1–M1 was the tract with the highest distance; and M1- and ParaC-Lenti presented a similar and intermediate distance.

After meeting the challenge of extracting reliable individual SM tracts by optimizing the settings of an automated tractography-based approach benefitting from the dHCP multi-shell dMRI data, we explored the WM maturational differences of low-risk preterm infants at TEA compared to full-term neonates. We then quantified the microstructure of each tract by extracting DTI and NODDI-derived diffusion metrics using a weighting approach which privileges the core of the tract and avoids potential bias linked to some inter-individual differences in tract volumes.

DTI and NODDI models present different trade-offs between complexity, biological plausibility, robustness, and run-time duration (Jelescu and Budde, 2017). Despite its widespread use in most studies of WM development (Dubois et al., 2014; Ouyang et al., 2019), DTI-derived metrics can be affected by several microstructural features and lack specificity to disentangle the complex properties of voxels containing crossing, kissing and fanning fibers (Zhang et al., 2012; Jeurissen et al., 2013). In contrast, NODDI allows a more sophisticated and biologically plausible multi-compartment model, relevant for developmental studies (Chang et al., 2015; Genc et al., 2017; Mah et al., 2017; Kimpton et al., 2021), but requires multi-shell data and increased processing time. Although potentially sub-optimal, we opted for default settings of diffusivities in the NODDI model, which were optimized for the adult WM but not for infants (Guerrero et al., 2019). In the absence of gold standards for infant-specific NODDI fitting to evaluate the metric maps, this was performed to maintain some consistency with previous studies (Guerrero et al., 2019; Fenchel et al., 2020).

Overall, the resulting metrics maps were consistent among PT and FT subjects, and metric differences across tracts were largely coherent between DTI and NODDI results in all infants. Once controlled for the effects of PMA at scan and global WM microstructure, we observed that AD, RD, MD decreased with GA at birth in all tracts, while FA, NDI and ODI increased. Globally, MD differences across tracts presented an opposite pattern to NDI (Kimpton et al., 2021), and differences in RD were highly similar to MD in all tracts, and opposite to FA in cortico-subcortical tracts, confirming that variations in MD and FA are likely largely driven by RD. Such diverse profiles across tracts might result from differences in both intrinsic microstructure (similar to adults) and in maturation (according to different myelination stages across tracts) (Dubois et al., 2014). During WM development, MD tends to decrease and NDI to increase with the growth of fibers and membranes, acting as barriers to the random water motion, while FA tends to increase, reflecting several factors including the presence of compact fiber tracts and increasing myelination (Beaulieu and Allen, 1994; Batalle et al., 2017; Kimpton et al., 2021). ODI describes the orientational dispersion of fibers within a tract, which is highly variable across tracts and likely changes during the growth and maturation of crossing fibers (Raghavan et al., 2021). Interestingly, in a few tracts, we did not observe significant relationship between FA or ODI (corrected for PMA at scan and WM residuals) and GA at birth, suggesting that both metrics might be less sensitive to detect subtle variations of microstructure in the settings of this study. Further analyses (beyond the scope of this study) could be performed to evaluate the potential correlations between all these metrics, as performed in previous studies (Kunz et al., 2014).

Moreover, the results of the maturational distance analyses underlined a high coherence between DTI sets which seemed fairly intuitive and in line with previous studies (Li et al., 2022). The differences observed between the DTI and NODDI sets (with NODDI set presenting more compact and roughly lower Mahalanobis distances values across tracts, with subtle variations in the main order of distances) suggested that NODDI metrics provide complementary information, probably due to their differing sensitivity to neurites growth and maturation. Thus, in line with previous studies (Batalle et al., 2017; Kimpton et al., 2021), our results highlighted the complementarity of these models and confirmed the relevance of NODDI-derived metrics for the study of WM microstructure maturation in the context of prematurity.

In contrast to the majority of previous studies, we focused on low-risk PT infants without overt brain abnormality at TEA, which is representative of the majority of children now born preterm in developed countries. Despite the presence of specific clinical risk factors in some of the PT infants (e.g., morbidities related to prematurity), the absence of a significant difference in the BSID-III outcome at 18 months of age between PT and FT infants corroborated that the included PT infants were at low-risk for neurodevelopment impairment.

In this study, we hypothesized that WM microstructure within SM tracts would show a significant maturational delay at TEA in PT infants compared to FT neonates, with distinct patterns as a function of GA at birth and across cortico-subcortical and cortico-cortical tracts. This was based on the assumption that early peripheral stimuli are essential for the emerging SM network maturation, and that preterm birth is associated with modified SM stimuli and experiences, notably related to numerous and various procedures in NICU (Mörelius et al., 2006; Gibbins et al., 2008), which might have a differential effect on somatosensory and motor systems (Duerden et al., 2018; Schneider et al., 2018; Jones et al., 2022). Of direct relevance, previous studies have reported either higher tactile sensitivity in PT infants at TEA (André et al., 2020), tactile hyporeactivity, lower brain responses (Maitre et al., 2017) and/or undifferentiated integration of nociceptive versus non-nociceptive stimuli (Fabrizi et al., 2011), in association with WM abnormalities (Brummelte et al., 2012; Zwicker et al., 2013).

Univariate dMRI approaches, based on individual derived metrics, are commonly applied to reveal WM developmental changes in the neonatal brain (Kunz et al., 2014; Ouyang et al., 2019). These analyses were thus used to evaluate the effects of several factors on the diffusion metrics measured in SM tracts.

Firstly, these were not related to the infants’ sex in our cohort. Combined with previous studies which showed inconsistent results (Pannek et al., 2014; Barnett et al., 2018; Kimpton et al., 2021), this observation suggests that sex effects may vary according to the studied brain regions. Surprisingly, we also observed no effect of main perinatal clinical risk factors (including preterm morbidities), despite numerous studies describing associations of WM abnormalities with obstetric, neonatal and postnatal factors (Pogribna et al., 2013; Brouwer et al., 2017; Barnett et al., 2018; Parikh et al., 2021), and with exposure to cumulative risk factors (Barnett et al., 2018). These negative results might be partly due to the specific study of low-risk PT infants.

In line with previous literature, we observed that diffusion metrics –even measured at TEA– were dependent on PMA at MRI (de Bruïne et al., 2011; van Pul et al., 2012; Kimpton et al., 2021) and were influenced by prematurity (Kunz et al., 2014; Kelly et al., 2016a; Batalle et al., 2017; Thompson et al., 2019; Dibble et al., 2021; Kimpton et al., 2021), with PT_EV infants showing more “immature” microstructural profiles (higher AD, RD, MD, lower FA, NDI, ODI) than PT_ML and FT infants. Interestingly, when group analyses were performed at the tract level, PT_ML showed no difference with the FT paired group, suggesting that the specific SM tracts studied here might not contribute significantly to the well-described whole-brain WM diffusion abnormalities in moderate-late PT (Kelly et al., 2016a; Thompson et al., 2019).

We then aimed to evaluate the maturational gap between PT infants and their full-term peers, based on SM tracts microstructural characteristics at TEA. Although univariate dMRI approaches allow some inference about the effects of prematurity on SM tract microstructure, they cannot reflect the inter-related complexity of processes involved in early brain maturation (Kostović et al., 2019), and are limited by the difficulties interpreting findings related to single metrics which are sensitive to different underlying microstructural properties and maturational processes. Also, quantifying the maturational degree across regions requires comparison of an infant’s data with a mature reference to account for “intrinsic” microstructural differences (Dubois et al., 2014). To overcome these limitations, we implemented an original multivariate approach already validated in neonatal and pediatric data (Kulikova et al., 2015; Dean et al., 2017; Li et al., 2022), that took advantage of the complementary information described by different DTI and NODDI metrics, to enable better characterization of SM tract maturation and the effects of prematurity as compared to typical development. Multivariate Mahalanobis distance was calculated in respect to a reference group (FT neonates) which provided typical values for the given tract. Importantly, this approach also allowed to take into consideration both the inherent variability of the diffusion metrics across tracts in the FT group and the correlations between these metrics. For each tract, the resulting maturational distance related to prematurity could be interpreted as a developmental gap between a PT infant at TEA and the FT control group.

Focusing on the PT_EV group, the comparison of distances across tracts highlighted the differential impact of prematurity on the SM tracts at TEA. For all sets of diffusion metrics, the impact increased in a caudo-rostral and central-to-peripheral manner following the typical progression of WM growth and myelination during infancy (Yakovlev and Lecours, 1967; Dubois et al., 2014) and within the CST tract (Kimpton et al., 2021), suggesting that early maturing tracts are less impacted. Furthermore, while this spatial pattern is globally consistent with previous studies of preterm infants (Wu et al., 2017; Knight et al., 2018), our results raise several interesting points regarding the functional role of the different SM tracts and related GM structures during development.

Firstly, we observed a differential impact of prematurity among tracts related to different cortical seeds. S1-subcortical tracts were systematically more impacted than M1- and ParaC-subcortical tracts. This suggests that S1 tracts may have a specific vulnerability to the deleterious effects of prematurity, possibly due to the altered SM perceptions and experiences as a result of early exposure to the ex-utero environment. Alternatively, the observed differences may reflect compensatory faster and/or more efficient maturational “catch-up” mechanisms in the M1/ParaC-subcortical tracts during the first post-natal weeks after preterm birth.

The similar profiles seen in the ParaC-subcortical tracts and related M1-subcortical tracts are less straightforward to interpret as the paracentral lobule includes both motor and somatosensory regions. Given the somatotopic organization of S1 and M1, this could suggest that connections related to the lower limb representations are less impacted by prematurity, which may be linked to the possible advanced maturation of these representations at early ages (Devisscher et al., 2021). A further possible explanation is that a greater number of motor fibers than sensory fibers were included in ParaC- tracts.

Secondly, the prematurity impact was also variable across sub-cortical related tracts. Of these, the cortico-Brainstem tracts appeared the least impacted. As these mainly correspond of CST fibers that myelinate early, notably at the level of the PLIC (Dubois et al., 2014; Kulikova et al., 2015; Kimpton et al., 2021), this is consistent with connections that have more advanced maturation at the time of birth being less vulnerable to prematurity (Wu et al., 2017). In agreement with the acknowledged vulnerability of thalamocortical connections following preterm birth (Ball et al., 2013a), the cortico-Thalamic tracts showed higher impact than cortico-Brainstem tracts, giving an “intermediate” profile compared to other studied tracts. The specific functional role of the thalamus, with essential input and output projections to the different SM regions, might help to modulate this vulnerability compared to other sub-cortical structures (Duerden et al., 2018; Schneider et al., 2018). Similarly, the cortico-Caudate tracts showed “intermediate” profile. This might result from an interplay between the high vulnerability of the caudate nuclei to prematurity (Nosarti et al., 2014; Back, 2015; Loh et al., 2017) and the adverse effects on the major efferent projections (McClendon et al., 2014) compared with more “preserved” afferent connections (from SM cortices). The cortico-Lenticular tracts systematically presented the maturational distance profile with the greatest impact of prematurity, suggesting their specific vulnerability. In addition to the known structural consequences of preterm birth on BG growth (Loh et al., 2017, 2020), different hypotheses can be proposed to explain this specific profile, especially knowing the anatomo-functional particularities of these tracts. As the dissected tracts include both (afferent) cortico-putaminal and (efferent) pallido-cortical fibers, the observed alteration may involve both the input (putamen) and output (GPi) structures of the BG, which have different functions in cortico-BG loops. We hypothesize that maturation of the efferent pallido-cortical fibers is specifically altered by prematurity, with functional effects on information reaching SM cortices, which might secondarily induce alterations of the descending cortico-striatal and cortico-pallidal fibers.

Finally, we observed the highest impact of prematurity in the cortico-cortical tracts, suggesting a particular vulnerability of these rostral structures, in line with the well-described caudo-rostral maturational pattern. Inter-hemispheric tracts presented lower impact than intra-hemispheric S1–M1 tracts, in line with the late and protracted maturation of such short-range connections (Kostović et al., 2019).

While interesting, the results should be interpreted cautiously given the limitations of diffusion MRI and tractography with relation to the image spatial resolution and the size of neonatal structures, and the presence of crossing fibers notably at the level of the corona radiata. However, the rare high-quality of dHCP neonatal data, the use of HARDI acquisition, the consistency of the tract’s delineation and the multivariate approach allowed us to overcome, at least partially, these limitations. In the future, it would be interesting to further investigate whether the vulnerability of SM tracts to prematurity is stable over development or whether “catch-up” development is present for some tracts, either before or after TEA. This would require the longitudinal evaluation of maturational distances defined with similar settings.