Subtle microstructural alterations in white matter tracts involved in socio-emotional processing after very preterm birth

Highlights

• •We investigated FA along white matter tracts involved in socio-emotional processing using both a traditional and functional data analysis tractometry approach.
• •Very preterm born children showed reduced FA along a middle portion of the right SLF I and anterior portion of the left SLF II.
• •FA along portions of the left SLF I, right SLF II, and right IFO  was significantly correlated with parent-reported social responsivity solely in very preterm born children.

Abstract

Children born very preterm (VPT, < 32 weeks of gestation) have an increased risk of developing socio-emotional difficulties. Possible neural substrates for these socio-emotional difficulties are alterations in the structural connectivity of the social brain due to premature birth. The objective of the current study was to study microstructural white matter integrity in VPT versus full-term (FT) born school-aged children along twelve white matter tracts involved in socio-emotional processing. Diffusion MRI scans were obtained from a sample of 35 VPT and 38 FT 8-to-12-year-old children. Tractography was performed using TractSeg, a state-of-the-art neural network-based approach, which offers investigation of detailed tract profiles of fractional anisotropy (FA). Group differences in FA along the tracts were investigated using both a traditional and complementary functional data analysis approach. Exploratory correlations were performed between the Social Responsiveness Scale (SRS-2), a parent-report questionnaire assessing difficulties in social functioning, and FA along the tract. Both analyses showed significant reductions in FA for the VPT group along the middle portion of the right SLF I and an anterior portion of the left SLF II. These group differences possibly indicate altered white matter maturation due to premature birth and may contribute to altered functional connectivity in the Theory of Mind network which has been documented in earlier work with VPT samples. Apart from reduced social motivation in the VPT group, there were no significant group differences in reported social functioning, as assessed by SRS-2. We found that in the VPT group higher FA values in segments of the left SLF I and right SLF II were associated with better social functioning. Surprisingly, the opposite was found for segments in the right IFO, where higher FA values were associated with worse reported social functioning. Since no significant correlations were found for the FT group, this relationship may be specific for VPT children. The current study overcomes methodological limitations of previous studies by more accurately segmenting white matter tracts using constrained spherical deconvolution based tractography, by applying complementary tractometry analysis approaches to estimate changes in FA more accurately, and by investigating the FA profile along the three components of the SLF.

1. Introduction

Very preterm (VPT; < 32 weeks of gestation) birth represents five percent of births worldwide (Walani, 2020). While these children often face life-threatening events and early life adversity, the Neonatal Intensive Care Unit (NICU) discharge rate of VPT born children has increased over the years (Bell et al., 2022). However, the enhanced survival rate for this group is also associated with an increase in neonatal problems and long-term morbidity, including neurodevelopmental problems (Fenoglio et al., 2017, Montagna and Nosarti, 2016). Children born VPT have an increased risk of motor deficits, cognitive deficits, and academic difficulties (Allotey et al., 2018, de Kieviet et al., 2009, Ritter et al., 2013, Twilhaar et al., 2018, Wolke et al., 2019).

Another documented neurodevelopmental complication in children born VPT is atypical socio-emotional development. Studies show that children born VPT present greater social withdrawal, poorer social skills, poorer facial expression processing, and problems with peer relations (Ritchie et al., 2015, Tang et al., in preparation). Throughout the research literature, a consistent profile of socio-emotional outcomes has emerged. This profile, characterized by emotional and socialization difficulties, inattention, and a greater risk of internalizing problems (Linsell et al., 2019, Montagna and Nosarti, 2016), is dubbed the “preterm behavioral phenotype” (Johnson & Marlow, 2011). The atypical socio-emotional development documented in this group is likely caused by differences in social information processing (Montagna & Nosarti, 2016).

Socio-emotional processing is coordinated by a large neural network, referred to as the social brain, which consists of multiple regions such as the posterior superior temporal sulcus, amygdala, orbitofrontal cortex, temporal poles, fusiform gyrus, inferior parietal lobule, and posterior inferior frontal gyrus (Catani et al., 2013, Fenoglio et al., 2017, Yang et al., 2015). However, due to VPT birth, the structure of the brain and these social brain regions in particular may be altered (Montagna & Nosarti, 2016). The most common structural deficit in children born VPT is periventricular leukomalacia (PVL) (Volpe, 2009). However, even in VPT born children without PVL, atypical grey and white matter development occurs (Inder et al., 2023). Studies have shown altered gray matter development and cortical thickness in the socio-emotional network of the brain in children born VPT. Importantly, these alterations have been found at different ages and often persist until adulthood (Alexander et al., 2019, Chau et al., 2019, Giménez et al., 2006, Healy et al., 2013, Mürner-Lavanchy et al., 2014, Nosarti et al., 2008, Peterson et al., 2000, Schmitz-Koep et al., 2021). However, structural deficits in the nodes of the socio-emotional network are not the only neural substrate for atypical socio-emotional processing, as the white matter tracts connecting these nodes can be altered as well.

The white matter tracts connecting the social brain regions and postulated to be most involved in socio-emotional processing are the inferior longitudinal fasciculus (ILF), the inferior fronto-occipital fasciculus (IFO), the uncinate fasciculus (UF) and the superior longitudinal fasciculus (SLF) (Wang et al., 2018). For a visualization of these white matter tracts see Fig. 1. The ILF connects the ventral extrastriate regions and portions of the inferior parietal lobe to the hippocampus, the amygdala, and the anterior temporal lobe (Wang et al., 2018). The IFO starts in the ventral occipital cortex, continues towards the dorsal temporal cortex,then connects to the UF, and finally ends in the inferior frontal, medial prefrontal, and orbitofrontal cortex (Catani and Thiebaut de Schotten, 2008, Wang et al., 2018). Both the ILF and the IFO connect multiple regions involved in face processing and play a role in the discrimination of facial expressions (Unger et al., 2016, Wang et al., 2018). The UF connects the temporo-amygdala-orbitofrontal network (Catani et al., 2013) and is implicated in emotional regulation, social valuation, and Theory Of Mind (TOM) (Eden et al., 2015, Fenoglio et al., 2017). The SLF connects the frontal, temporal, and parietal cortex and has been implicated in empathy and TOM (Hamzei et al., 2016, Parkinson and Wheatley, 2014a, Wang et al., 2018). While not universally agreed upon, the SLF is often dissected into three components, namely SLF I, SLF II, and SLF III (for a full review of these different components and their anatomical definitions see Janelle et al., 2022).

Fig. 1.

The White Matter Tracts Postulated to be Involved in Socio-emotional Processing. Note: Visualization of the bilateral uncinate fasciculus (green), inferior longitudinal fasciculus (blue), inferior fronto-occipital fasciculus (red), superior longitudinal fasciculus (yellow), superior longitudinal fasciculus II (grey), and superior longitudinal fasciculus III (brown) in a glass brain. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Several studies have investigated the microstructure of these white matter tracts involved in socio-emotional processing, comparing VPT to full-term (FT) born children and adolescents using Diffusion Tensor Imaging (DTI). These studies mainly investigate the fractional anisotropy (FA) of these tracts; FA is a scalar measure between 0 and 1 reflecting the anisotropy of diffusion (Curran et al., 2016). It is often used as an index of white matter integrity (higher FA reflecting higher integrity) due to its correlations with axon count, axon density and degree of myelination, although this is an oversimplification (Curran et al., 2016). Studies in samples with VPT children consistently show a decrease in FA for the UF (Constable et al., 2008, Mullen et al., 2011, Travis et al., 2015, Vollmer et al., 2017, Young et al., 2018). However, findings regarding the ILF, IFO, and SLF are mixed (Constable et al., 2008, Dodson et al., 2017, Feldman et al., 2012aa; Kallankari et al., 2023, Salvan et al., 2014, Travis et al., 2015, Vollmer et al., 2017, Young et al., 2018). This variety in results can partially be explained by the variety in methods used in these studies.

Most prior studies (Constable et al., 2008, Feldman et al., 2012aa; Mullen et al., 2011, Vollmer et al., 2017, Young et al., 2018) made use of voxel-based morphometry or Tract Based Spatial Statistics (TBSS) (Smith et al., 2006). However, the interpretation of findings using these methods is complicated by anatomical misalignment, choice of registration target, and the inability of the FA skeleton to accurately represent white matter tracts (Bach et al., 2014, Van Hecke et al., 2016, Yeatman et al., 2012). Other studies use well-validated automated tractography methods to overcome these limitations. These studies typically average diffusion properties over the entire length of the tract. However, as demonstrated by the so-called tractometry approach of Yeatman et al (2012), diffusion properties largely vary along the tract, FA values increase and decrease at specific regions in the tract during development, and group differences in FA may only be present at specific parts of the tract and not along the entire tract (Yeatman et al., 2012). Against this background it is eligible to investigate FA profiles along the tracts (Chamberland et al., 2019, Colby et al., 2012, Jones et al., 2005, Yeatman et al., 2012, Zhang et al., 2022), especially since previous research in VPT samples has shown complex profiles of both increased and decreased FA compared to a control group in the right IFO (Travis et al., 2015). Previous tractometry analyses in VPT samples have been performed using automated fiber quantification (AFQ; Dodson et al., 2017, Travis et al., 2015, Yeatman et al., 2012), but recently more accurate methods have been developed for both the automated tractography and tractometry analysis.

In the current study we use TractSeg, a state-of-the-art convolutional neural network-based approach that directly, automatically, and accurately segments white matter tracts based on fiber orientation density peaks (Wasserthal et al., 2018, Wasserthal et al., 2020). TractSeg is able to more accurately segment white matter tracts than other automatic segmentation methods. Moreover, TractSeg allows to reconstruct the three distinct components of the SLF separately; in contrast to previous studies performed in VPT samples where the SLF was studied as a single entity or split into a ventral and dorsal component. Most importantly, TractSeg allows to analyze FA values along multiple segments across the whole tract by using an approach based on the Bundle Analytics (BUAN) framework of Chandio et al. (2020). Finally, we combine both a traditional tractometry analysis with a functional data analysis approach, to overcome statistical limitations and evaluate the robustness of our findings. Thus, we believe the current study is one of the most accurate investigations of FA differences between VPT and FT groups.

The current diffusion MRI (dMRI) study is part of a larger research project to study the socio-communicative and socio-emotional vulnerabilities seen in school-aged children born prematurely, compared to age- and sex-matched FT peers. We will study differences between the VPT and FT groups in the FA tract profiles of the major tracts involved in socio-emotional processing, i.e., the bilateral ILF, IFO, UF, and SLF I-II-III. Next, we will compare socio-emotional functioning between the VPT and FT children by investigating group differences on the Social Responsiveness Scale (SRS-2), and we will conduct exploratory correlational analyses, separately for the VPT and FT groups, examining the relationship between FA values along these tracts and SRS-2 scores. This investigation aims to further explore the connection between white matter microstructure in social brain tracts and potential variation in reported social behavior. Based on earlier literature, we hypothesize a reduction in FA value for the ILF, IFO, UF, and SLF (I, II, III) in children born VPT compared to the FT group.

2. Material and methods

2.1. Participants

The sample of the current study consisted of two groups of age-matched FT and VPT born children. VPT participants were recruited based on the database of the NICUof the University Hospital in Leuven. FT participants were recruited by distributing brochures to schools or youth organizations. Inclusion criteria for the VPT and FT group were (1) an age range between 8 and 12 years, (2) an intelligence quotient (IQ) of at least 70, and (3) birth between 24 and 32 weeks of gestational age for the VPT participants and birth at 37 weeks or later for the FT participants. Participants were excluded if they had significant hearing or vision impairments, were a non-Dutch native speaker, had a history of any neurological disorder, had any contraindication for MRI research, or if they had a formal diagnosis of any psychiatric disorder, with special attention to autism spectrum disorder (ASD) as defined by Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Text Revision (DSM-5-TR) criteria (American Psychiatric Association Publishing, 2022). After recruitment, our sample consisted of 40 VPT born children and 38 FT born controls. However, two VPT participants had to be excluded because no dMRI data was available for them, and three because of low data quality due to movement in the scanner, yielding a final sample of 35 VPT born children and 38 FT born peers.

2.2. Clinical and intelligence measures

For all VPT participants, we collected clinical information of their stay in the NICU, i.e., gestational age, birth weight, Apgar scores (after one and five minutes), intracerebral hemorrhage, PVL, days of oxygen therapy, and duration of stay. Information about gestational age and birthweight was collected for the FT children as well. Quotients of performance (PIQ; Block design and Figure puzzles) and verbal (VIQ; Similarities and Vocabulary) intelligence were collected for all participants at the moment of visit using an abbreviated version of the WISC-V (WISC-V-NL | Wechsler Intelligence Scale for Children-V, 2018) with the norms for Dutch-speaking children.

2.3. Assessment of difficulties in social functioning

The Social Responsiveness Scale (Constantino, 2012; SRS-2) was used to measure difficulties in parent-reported social behavior and the presence of quantitative autism characteristics. The outcome of this instrument consists of a total score and scores on five subscales: social awareness, social cognition, social motivation, social communication, and restricted interests. The SRS-2 yields both raw scores and normalized T-scores. The total T-score expresses the deficit in social functioning in interaction with others, thus higher scores correspond with more severe dysfunction. Scores between 61 and 75 correspond to a mild social deficit in social behavior, while scores of 76 or more indicate a clinically significant deficit in social functioning. The subscales are not used for clinical decision making but allow to further pinpoint the symptomatology. The social awareness scale gives an indication of an individual’s ability to recognize social cues of other persons; the social cognition scale assesses the interpretation of social behavior; the social motivation scale measures how motivated the individual is to participate in social interaction; the social communication subscale assesses interactive communication with others in social situations; and the restricted interests scale measures stereotypy in behavior and specific interests. The norms for Flemish children and parent ratings were used.

2.4. Diffusion MRI

2.4.1. Acquisition

Acquisition of the diffusion weighted imaging (DWI) data was performed using a 3T Philips Achieva System at the university hospital in Leuven (Belgium) using a 32-channel head coil. We used two spin-echo echo-planar imaging (EPI) sequences with one frontal and one reverse phase encoding gradient. The frontal sequence had the following parameters: 58 continuous sagittal slices, slice thickness = 2.5 mm, repetition time (TR) = 4 s, echo time (TE) = 0.082 s, interslice gap = 0, field of view (FOV) = 200 X 240 X 145, matrix size = 96 X 96 X 96, SENSE acceleration factor = 2, b0 = 1 vol, flip angle = 90°, b-value = 1300 s/mm², phase coding direction = anterior > posterior, and directions = 61. The reverse sequence had the following parameters: 58 continuous sagittal slices, slice thickness = 2.5 mm, repetition time (TR) = 4 s, echo time (TE) = 0.078 s, interslice gap = 0, flip angle = 90°, field of view (FOV) = 200 X 240 X 145, matrix size = 96 X 96 X 96, SENSE acceleration factor = 2, b0 = 1 vol, b-value = 1000 s/mm², phase encoding direction = posterior > anterior, and directions = 7.

2.4.2. Pre-processing

After the DWI data were collected and saved, they were converted from PAR/REC format to NIFTI using dcm2niix (Li et al., 2016). The pre-processing of the DWI data was done using a combination of the Mrtrix3 (Tournier et al., 2019), FSL (Jenkinson et al., 2012) and ANT’s (Avants et al., 2014) software packages. First, all images of the frontal sequence were denoised using the MP-PCA algorithm (Veraart et al., 2016a, Veraart et al., 2016b). Second, correction for Gibbs ringing artefacts was performed. Third, images were corrected for eddy currents, susceptibility distortions (using the average b0 images of the frontal and the reverse sequence), and subject motion using the TOPUP and EDDY toolboxes from FSL (Andersson et al., 2003, Andersson and Sotiropoulos, 2016). Fourth, images were corrected for B1 field inhomogeneity using the N4biasfield algorithm of the ANT’s toolbox (Tustison et al., 2010). Fifth, a brain mask was calculated for each participant using FSL’s BET (Smith, 2002). Sixth, images were rigidly registered to the MNI template from the TractSeg resource folder using FSL’s FLIRT, and the B-vectors were rotated accordingly (Jenkinson et al., 2002, Jenkinson and Smith, 2001). Finally, FA values for each participant were calculated using FSL’s DTIFIT (Hernández et al., 2013).

2.4.3. Quality control of DWI data

After pre-processing, the quality of the DWI data was checked. First, to check if the conversion from PAR/REC to NIFTI was successful, the first eigenvector of diffusion was visually plotted using FSLEYES. The anatomical plausibility of the direction of this eigenvector was verified and was confirmed to align with the anterior corpus callosum for all participants.

Further quality control was performed both at the individual level and at the study level using FSL’s EDDY QUality Assessment for dMRI (QUAD) and Study-wise QUality Assessment for dMRI (SQUAD), respectively (Bastiani et al., 2019). This resulted in quality indices for volume-to-volume motion, within-volume motion, eddy current-induced distortions and susceptibility-induced distortions (see Supplementary Material for the complete SQUAD report and an analysis of group differences on the SQUAD quality indices). Participants with an average translation of 2.5 mm or more, and/or rotation of 1.5° or more were excluded from the data. Based on these criteria one VPT participant had to be excluded. Visual inspection of the report indicated another two outliers in absolute movement in the VPT group. These VPT participants were also excluded from further analysis.

2.4.4. Bundle segmentation

To segment our white matter bundles of interest we used TractSeg (Wasserthal et al., 2018). This method makes use of an encoder-decoder fully connected convolutional neural network (CNN) that processes the extracted fiber orientation density (FOD) peaks and then generates tract probability maps which are used to segment accurate bundle-specific tractograms.

To extract the principal FOD peaks as the input for TractSeg, the following steps were performed. First, response functions were obtained for each participant’s data using Mrtrix3 and the Tournier algorithm (Tournier et al., 2013). Next, FOD’s were estimated using single-shell single-tissue constrained spherical deconvolution (Tournier et al., 2007). The first three principal FOD peaks were extracted and used for further analysis in TractSeg to perform bundle segmentation, segmentation of the begin and endpoints of the tracts, and to create tract orientation maps (Wasserthal et al., 2018). For each tract, probabilistic tractography was performed using the tract orientation maps and a large number of fibers (N = 5000) was generated to reduce variability.

FA values along these tracts were obtained based on the BUAN approach of Chandio et al. (2020). BUAN samples the streamlines of each white matter tract into an equal number of equidistant segments (n = 100) and then calculates the centroid (i.e., the middle of all streamlines of the tract), which is then also sampled in 100 segments. Then for each streamline all streamline segments were assigned to the closest centroid segment, forming a tract segment. For each tract segment the average FA of all streamlines was calculated, resulting in 100 FA measurements per tract. This was done separately for each white matter tract of interest on both the left and right side of the brain.

2.5. Statistical analysis

Demographic differences between the groups were investigated using student’s t-tests and chi-square tests. Group differences on the SRS-2 scores were investigated using student’s t-tests. The significance level was set at α = 0.05.

For the analysis of group differences in FA along the tract profiles we first employed the traditional approach by Yeatman et al. (2012), adapted by Wasserthal et al. (2020). While this approach is often used, it entails potential issues regarding multiple comparisons, non-gaussian distributions and spatial dependencies (for an extensive overview, see Muncy et al., 2022). Therefore, we also performed a complementary functional data analysis (FDA) approach based on Chandio (2022) and Goldsmith & Kitago (2016) to adress these issues.

Using the traditional approach of Wasserthal et al. (2020), a linear model was fitted for each of the 100 tract segments of a particular tract and a non-parametric permutation-based two sample t-test with 5000 iterations was performed to investigate group differences between the VPT and FT groups. Controlling for covariates was done by regressing them out of the data before the t-test was calculated. To ensure that group differences in head movement would not drive group differences in the results (Yendiki et al., 2014), subject motion was regressed out using the average voxel displacement in comparison to a reference slice (i.e., absolute motion). Age was also added as a nuisance parameter as previous studies have shown that tract profiles change with age (Yeatman et al., 2012). Finally, sex was added as a covariate, since substantial sex differences in FA values along tracts have been documented (Kanaan et al., 2012).

We used the single threshold test variant of this procedure to account for multiple comparisons between the adjacent segments of the tract (for more details, see Nichols & Holmes, 2002). This offers a distinct corrected $a$-FWE value for each separate tract, and group comparisons for a particular segment are deemed significant if the corrected p values are lower than this $a$-FWE. Consistent with comparable studies (Dodson et al., 2017, Muncy et al., 2022, Nabulsi et al., 2023, Sang et al., 2022, Travis et al., 2015, Wasserthal et al., 2020) and due to the fact that the selection of our tracts was hypothesis-driven (Chamberland et al., 2019), we did not additionally control for multiple comparisons between the different white matter tracts. Note, that the $a$-FWE value for each separate tract was comparable to the recommended cutoff of 0.001 provided using the guidelines of Chandio et al. (2020) for comparing 30 white matter tracts using the bounds of correlated tests (Das & Bhandari, 2021). The raw p values and results including a correction for multiple comparison between the white matter tracts can be found in Supplementary Materials Fig. S1.

A complementary analysis of the same tracts and tract segments was performed using function-on-scalar (FOSR) regression, a variant of FDA (Ramsay & Silverman, 2005). FDA serves as a comprehensive framework for analyzing data distributed along a continuum or surface, operating under the assumption that an underlying function gives rise to the observed data. This allows us to investigate tract profiles as single entities instead of a series of observations or segments (Chandio, 2022, Goldsmith et al., 2011, Muncy et al., 2022, Sørensen et al., 2013). This method has previously been used successfully to investigate group differences in FA along white matter tracts (Chandio, 2022, Goldsmith et al., 2011, Goldsmith et al., 2012, Muncy et al., 2022, Verde et al., 2014, Zhu et al., 2011).

To implement this analysis, we adopted the variational Bayesian implementation by Goldsmith and Kitago (2016) for fitting a linear FOSR model, following the methodology of Chandio (2022). The function of the FA values of the 100 segments along the tract for every participant was used as response variable with group as the predictor of interest, and age and sex as scalar covariates. The model was fitted using the refund package in R. FOSR plots for each white matter tract were generated, with the segments along the tract as the x-axis and b1 (representing group differences) on the y-axis. In line with Chandio (2022) and more conservative than Muncy et al. (2022), we opted to define significant segments as those exhibiting deviations of ± 0.02 (point estimate) from zero slope in b1(t) along the tract. However, recognizing the limitations of the frequentist method for parameter estimation, we further refined our approach by using a variational Bayesian implementation (Goldsmith & Kitago, 2016). To identify credible group differences, we employed a region of practical equivalence (ROPE) procedure (Kruschke, 2013, Kruschke, 2015). Specifically, we classified regions along the tracts where the 95% Bayesian credible interval exceeded ± 0.02 as credible group differences in FA. This Bayesian FOSR approach not only offers a more robust foundation for parameter estimation, but also enhances our ability to discern meaningful variations along the tracts with greater confidence.

To investigate the correlation between the total T-scores on the SRS-2 and FA value along the tract, the traditional approach of Wasserthal et al. (2020) was used to perform a parametric permutation-based Pearson correlation with 1000 iterations. This was done for the two groups separately. The same single threshold test method as before was used to correct for multiple comparisons within a tract.

3. Results

3.1. Participant demographics

Our final sample consisted of 38 FT and 35 VPT born children. Demographic data, test statistics and p-values for both groups can be found in Table 1. The groups were well-matched in terms of age and sex. By definition, they differed highly significantly in terms of gestational age and birthweight. VIQ and PIQ were average in the VPT group, but significantly lower than the FT group. Clinical characteristics from their stay at the NICU were collected for all VPT participants. Following the guidelines of the World Health Organization (WHO), 15 had low birth weight (LBW; <2500 g), with 12 classified as very low birth weight (VLBW; <1500 g), and 8 falling into the extremely low birth weight category (ELBW; <1000 g). Mean Apgar score after one and five minutes was 6.9 (SD = 2.2) and 8.4 (SD = 1.1), respectively. The mean number of days that oxygen therapy was given was 20.1 days (SD = 24.9). Intraventricular hemorrhage (IVH) was documented (according to Papile et al., 1978) in two children (5.7%), one of them had IVH grade one, the other IVH grade two. None of the VPT children had documented PVL. Finally, the average duration of stay in the NICU was 58 days (SD = 36).

Table 1.

Demographics: Full-term and very preterm group.

	Very preterm (n = 35) Mean (SD)	Full-term (n = 38) Mean (SD)	t (71) / $X^2$ (1)	p
Age (years)	9.7 (1.4)	9.8 (1.3)	0.147	0.884
Sex (% male)	80 %	79 %	0.012	0.912
Gestational age (weeks)	28.7 (2.1)	40.2 (1.3)	27.929	<0.001
Birth weight (g)	1328 (4 2 4)	3685 (4 2 2)	23.773	<0.001
Verbal IQ	101 (17)	118 (12)	4.851	<0.001
Performance IQ	96 (11)	107 (11)	4.051	<0.001

Note: Means, student’s t-statistics, chi square tests, and p-values for the demographic data of both groups.

3.2. Group differences in FA

Fig. 2 shows an overview of the tract profiles for the two groups for the white matter tracts of interest using the traditional approach along with the corresponding $a$-FWE values, t-statistics and minimum p-values. A more detailed figure with the corresponding p-value for each of the 100 segments in the tract instead of just the minimum p-value can be found in the Supplementary Materials Fig. S1. We found significant group differences in FA values in the right SLF I ($a$-FWE = 0.001847, t (71) = 4.075055, minimum p-value = 0.000118) and left SLF II $(a$-FWE = 0.001335, t (71) = 3.493966, minimum p =.000824). In the right SLF I, there was a portion with significantly decreased FA values for the VPT group in the middle of the tract (see Fig. 3A). For the left SLF II we found significantly decreased FA values for the VPT group in an anterior segment (see Fig. 3B). There were no significant group differences in FA values along the bilateral IFO, ILF, SLF III, and UF, and along left SLF I and right SLF II.

Fig. 2.

FA tract profiles of the 12 white matter bundles for the VPT and FT group. Note: The mean FA profile for each tract is depicted for the VPT group (orange) and FT group (blue). The X-axis represents the position along the tracts, which are all divided in 100 segments according to the BUAN approach. The Y-axis represents the FA value for a particular segment. Shaded orange and blue regions indicate the 95% confidence interval of FA for each group. The red dotted vertical lines indicate all positions along the bundle where there is a significant group difference in FA value. Under each bundle, the value of the t-test statistic, minimum p-value and $a$-FWE value are displayed. IFO = inferior fronto-occipital fasciculus, UF = uncinate fasciculus, ILF = inferior longitudinal fasciculus SLF = superior longitudinal fasciculus. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3.

Portions of the right SLF I and left SLF II with significantly reduced FA in the VPT group. Note: Visualization of the right SLF I (A), left SLF II (B) and left as segmented via TractSeg. Portions of the tracts with significant reduction in FA for the VPT group are indicated in red. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The analysis using Bayesian FOSR confirmed the abovementioned findings by showing credible decreases in FA for the VPT group along the middle part of the right SLF I and the anterior part of left SLF II, see Fig. 4. In addition, a credible decrease in FA for the VPT group was also found at the anterior end of the left UF. Moreover, credible increases for the VPT group were found along an anterior part of the right SLF I, and anterior part of the left IFO. A more detailed figure of all white matter tracts of interest can be found in the Supplementary Materials Fig. S2.

Fig. 4.

FOSR Plots of White matter Tracts with Credible Group Differences. Note: The top image represents the mean estimated FA profile along the tract (b0) for both groups. The bottom image represents the group differences in FA along the white matter tracts b1(t), with positive values indicating increased FA for the VPT group and negative values indicating decreased FA for the VPT group. Thick colored lines represent the mean of the posterior, the thin colored lines represent the 95% Bayesian credible interval for each tract. The horizontal red dotted lines, represent the region of practical equivalence. In both images the x axis represents the location (t) along the white matter tracts. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.3. SRS-2 group differences and clinical cut-offs

Table 2 provides summary statistics for both groups on parent-reported social difficulties, as assessed by SRS-2. While the VPT group scored higher than the FT group on the mean total T-score, the group difference was not significant (t (71) = -1.630, p =.109, d = -0.38, 95 % CI [-0.84, 0.08]). No significant differences between the two groups were found for any of the subscales (all p >.05), except for difficulties with social motivation where the VPT group scored significantly higher than the FT group (t (71) = -3.21, p =.002, d = -0.77, 95 % CI [-1.24, −0.29]). Based on the clinical cut-off of the SRS-2, two children in the VPT group were classified as “severely disturbed in social responsivity”, while three children in the FT group were classified with a “mild disturbance in social responsivity”. All other children scored in the normal range.

Table 2.

Mean scores on parent-reported social responsivity for VPT and FT groups.

Scales	Very preterm (n = 35)Mean (SD)	Full-term (n = 38)Mean (SD)	t (71)	p	Cohens d
Total	51.57 (12.94)	47.85 (7.56)	−1.630	0.109	−0.38
Social Awareness	49.77 (10.85)	50.47 (8.27)	0.312	0.760	0.07
Social Cognition	51.23 (11.54)	46.87 (7.80)	−1.904	0.061	−0.45
Social Communication	50.89 (11.43)	48.61 (8.17)	−0.986	0.327	−0.23
Social Motivation	52.97 (11.31)	45.90 (6.79)	−3.271	0.002	−0.77
Restricted interests	52.49 (15.13)	48.66 (7.72)	−1.377	0.173	−0.32

Note: Means, student’s t-statistics, p values and effect sizes for the T-scores of all scales of the SRS-2.

3.4. Correlations between FA and SRS-2

In the VPT group, there was a significant positive correlation [r (33) = 0.56, $a$-FWE = 0.000761, minimum p =.000473] in a posterior part of the right IFO (Fig. 5A), and significant negative correlation [r (33) = -0.55, $a$-FWE = 0.000921, minimum p =.000713] in two portions of the right SLF II (Fig. 5B) and a small middle portion of the left SLF I (Fig. 5C) [r (33) = -0.52, $a$-FWE = 0.001494, minimum p =.001456]. None of the other tracts had a significant correlation with SRS-2 scores. For the FT group, no significant correlations were found between FA values along any of the tracts and the total T-score on the SRS-2.

Fig. 5.

Significant correlations between FA along the tract and total SRS-2 T-scores for the VPT group. Note: Visualization of the right IFO (A), right SLF II (B) and left SLF I (C) of the VPT group as segmented via TractSeg. Portions of the tracts with significant correlations between FA and the total SRS-2 scores are indicated in red. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Discussion

Using TractSeg, we reconstructed 12 white matter tracts related to socio-emotional processing in a sample of school-aged children born VPT and FT. Profiles of FA along the tracts were compared between the two groups using both a traditional and FDA analysis approach. To further investigate the association between white matter microstructure in these tracts and reported deficits in social behavior, exploratory correlations were performed for both groups between FA values along the white matter tracts of interests and total T-scores on the SRS-2.

Both the traditional and FDA approach showed significant differences between the two groups at the same portions of the right SLF I and left SLF II. Similarly, the Bayesian FOSR also showed credible differences along an additional segment of the right SLF I and along additional segments of both the left UF and IFO (although smaller than those in the SLF I, II). These differences were not significant using the traditional approach, although we can see statistical trends at the same regions (see Supplementary Material Fig. S1). In accordance with recent reports, the FDA approach seems to be a more valid and sensitive approach to perform statistical analysis on tractometry data (Chandio, 2022, Goldsmith et al., 2012, Muncy et al., 2022). Nevertheless, by employing these two complementary methodologies, we successfully mirrored significant differences in the right SLF I and left SLF II. Although these differences in FA may be subtle, we provide converging evidence for its existence. Thus, for the remainder of the discussion, we will only consider and interpret the overlapping findings of both approaches.

The reduction in FA along the middle part of the right SLF I and anterior part of the left SLF II likely indicates decreased axonal myelination or less densely packed axons in these regions for the VPT group (Beaulieu, 2002, Curran et al., 2016, Jones et al., 2013). Myelination of white matter tracts is performed by oligodendrocytes, which develop from pre-myelinating cells (Back et al., 2007;Volpe, 2011). These pre-myelinating cells are vulnerable to hypoxia–ischemia and systemic inflammation which are common in VPT birth, so damage to these cells can contribute to altered myelination in children born VPT (Back et al., 2007; Volpe, 2011). While this has already been well characterized in periventricular tracts (Dodson et al., 2017, Travis et al., 2015), the current study adds to the limited findings regarding altered white matter microstructure in non-periventricular tracts and regions (Travis et al., 2015). Documented group differences between VPT and PT populations in FA can either reflect long term consequences of early brain injury, altered maturation, or developmental experiences driving white matter development. However, the fact that our sample of VPT children had a relatively low incidence of NICU complications, with none of the VPT participants having documented PVL and only two participants with grade 1 and 2 IVH, and that the SLF is one of the last white matter tracts to mature (Zhang et al., 2007), suggest that the observed differences here reflect altered white matter maturation due to premature birth (Grotheer et al., 2023). Contrary to previous studies (Constable et al., 2008, Dodson et al., 2017, Mullen et al., 2011, Salvan et al., 2014, Travis et al., 2015, Vollmer et al., 2017, Young et al., 2018) we did not find any significant differences in the bilateral ILF, UF, and IFO. The lack of significant differences in these tracts can be explained by the few neonatal complications in our sample resulting in relatively normal brain development. Moreover it could be that differences in these white matter tracts may already have been resolved or only emerge at a later age (Dodson et al., 2017, Feldman et al., 2012aa; Travis et al., 2015, Yeatman et al., 2012).

This was the first study in a preterm sample to study the FA profile of the three components of the SLF. While Kallankari et al. (2023) also investigated FA in the three components of the SLF in a sample of 9-year-old VPT children, they investigated differences in the mean FA of these three components. Similar to our findings, they found reduced FA in the right SLF I in the VPT group. Yet, they found no significant differences in any of the other components. Here, we corroborate and extend their findings of reduced average FA in the right SLF I by specifically demonstrating reduced FA in the middle portion of this tract.

The SLF as a whole has been implicated in empathy and TOM (Hamzei et al., 2016, Parkinson and Wheatley, 2014a, Wang et al., 2018), but the distinct roles of the three separate components in socio-emotional processing are scarcely investigated. There is neuroanatomical evidence that both the SLF II and SLF III may play a role in social learning but in a different manner (Hecht et al., 2015). Namely, the SLF II is involved in action observation while the SLF III is involved in imitation and higher order cognition, both of which are necessary for social learning (Hecht et al., 2015). The SLF I, in particular, connects the superior parietal and superior frontal lobes to the dorsal premotor and dorsolateral prefrontal regions (Makris et al., 2005, Wassermann et al., 2016). The dorsal premotor cortex is one of the areas which hosts motor neurons, which play a huge role in social cognition and social interactions, providing a possible role for the SLF I in socio-emotional processing by being involved in embodied social cognition (Bonini et al., 2022, Comes-Fayos et al., 2018, Wang et al., 2018).

Aside from neuroanatomical evidence implicating the SLF in socio-emotional processing, there is also evidence from psychiatric and healthy populations. Altered FA in the left and right SLF has commonly been reported in ASD, a neurodevelopmental disorder characterized by deficits in social communication and social interaction. Studies in ASD samples found both increased and decreased FA (for an overview, see Fitzgerald et al., 2018). Moreover, lower FA values in the left SLF in ASD have been associated with reduced social cognition (Im et al., 2018). Conversely, higher FA in the bilateral SLF of healthy adults was positively associated with empathy (Parkinson and Wheatley, 2014b). Finally, a large number of studies found that the SLF is also associated with emotion recognition abilities (Wang et al., 2018).

Only a limited number of studies with VPT born children have investigated the relationship between white matter microstructure and measures of socio-emotional processing. They found a negative correlation between FA in the UF and emotion moderation scores (Kanel et al., 2021), a negative correlation between FA in the bilateral IFO and SLF with internalizing symptoms (Loe et al., 2013), a negative correlation between FA in the right IFO and externalizing symptoms (Loe et al., 2013), and a negative correlation between fiber density in the bilateral UF, IFO, left SLF I and right SLF II and III and externalizing symptoms (Gilchrist et al., 2023). The current study adds to the limited literature on this topic.

There were no differences between the two groups in reported difficulties in social functioning as assessed by the total SRS-2 T-scores. However, the VPT group did score significantly higher on the social motivation subscale, thereby indexing slightly more difficulties with social motivation as compared to their FT peers. In general, both groups scored very similar to the general population reference group. Based on the clinical cut-off of the SRS-2, two children in the VPT group were classified as severely disturbed in terms of social responsivity, and three children in the FT group were classified with mild disturbance in social responsivity. Our findings showed that in the VPT group higher FA values in segments of the left SLF I and right SLF II were associated with better social functioning (i.e., lower total T-scores on the SRS-2). Surprisingly, the opposite was found for segments in the right IFO where higher FA values were associated with worse social functioning (i.e., higher total T-scores). This may seem puzzling at first. However, increased FA does not necessarily point to increased white matter integrity or increased maturity, but can simply reflect a reduced number of crossing fibers or thinner axons (Dodson et al., 2017).

The tracts and portions with significant correlations in our study did not overlap with the observed group differences, and significant correlations were only found for the VPT group. This seem contradictory, however this pattern of findings has emerged in multiple studies in VPT samples investigating group differences and associations between FA and behavioral, cognitive, or developmental outcomes (Brenner et al., 2021, Dodson et al., 2017, Feldman et al., 2012aa; Feldman et al., 2012b, Kelly et al., 2021, Loe et al., 2013, Vollmer et al., 2017). It is important to take into account here that FA is influenced by a multitude of physiological properties (e.g., crossing fibers, axon density, myelin density). Thus, for VPT born children who may have experienced brain injury and white matter dysmaturity, the processes driving white matter changes may be different as compared to the FT group (Loe et al., 2013, Travis et al., 2016). Variation in FA in these segments could represent different physiological processes or balances for the VPT and FT group (Feldman et al., 2012a; Feldman et al., 2012b, Travis et al., 2016). Thus, different physiological properties or a different balance between them (Travis et al., 2016) could be associated with social responsivity. Nonetheless, showing relations between white matter microstructure and measures of socio-emotional development yields promise for predicting atypical (social-emotional) development or psychiatric conditions from connectivity metrics such as tractometry (Chamberland et al., 2021, Kanel et al., 2021, Neher et al., 2024), eventually creating possibilities for interventions.

While the relations between the findings from our study are complex, they do align with previous work in VPT samples. Mossad et al., 2020, Mossad et al., 2021 followed a cohort of VPT born children at different timepoints from birth until eight years of age, and investigated a variety of measures tapping into socio-emotional development. While at 8 years of age, they did not find any significant differences compared to FT controls on the SRS-2 subscales, the VPT sample did perform significantly worse in a TOM social attribution task, and rated emotional faces significantly different. Moreover, the VPT group showed altered functional connectivity in both the face processing and the TOM network. Thus, while parent reported social responsivity did not differ, social information processing at both the behavioral and neural level did differ between the groups. The same may hold true for our study. Our findings nicely align with those of Mossad et al. (2021), since delayed development of the SLF involved in TOM can contribute to the altered functional connectivity in the TOM network.

5. Limitations

Our study focused only on FA as a measure of white matter integrity, as this is the most important and most used metric in the literature. Additional diffusion metrics, such as MD, AD and RD, could provide a more comprehensive assessment of white matter microstructure. Moreover, the interpretation of the applied tensor model is complicated by its inability to model regions with crossing fibers and its lack of specificity to distinguish between different biological outcomes. Future studies may use high angular resolution diffusion weighted imaging (HARDI) and higher order diffusion models which are able to model crossing fibers and provide more specific measures of white matter microstructure along these white matter tracts. Another limitation is the modest sample size, a larger sample would enhance the ability to detect more subtle group differences and associations. A final limitation is the relatively wide age range among our participants. As mentioned before, age is an important confound, because FA increases until adulthood (Yap et al., 2013) and FA changes at specific places in the tracts at a specific time in development (Yeatman et al., 2012). This implies that even with a four-year age range among participants, there could have been a developmental effect. We tried to control for this as best as possible by including age as a nuisance parameter. A solution for future research would be longitudinal studies in which the same group of children would be studied at multiple different timepoints across development.

6. Conclusion

In conclusion, we investigated FA profiles of white matter tracts related to socio-emotional processing in a sample of school-aged VPT and FT born children using TractSeg. We performed both a traditional and functional data analysis approach of tractometry analysis, with both methods finding reduced FA for the VPT group along the middle portion of right SLF I and the anterior portion of the left SLF II. These group differences may reflect altered white matter maturation due to premature birth and may contribute to altered functional connectivity in the TOM network which has been documented in earlier work with VPT samples. Reported difficulties in social functioning for the VPT group were negatively associated to FA values in portions of left SLF I and right SLF II, and positively associated to FA values in a posterior region of the right IFO. Future studies should apply a longitudinal design and use higher order diffusion models to overcome the limitations of the DTI framework and to more accurately investigate microstructural alterations along these white matter tracts and to investigate how these change with age.

Funding

TT is supported by the Fund Child Hospital UZ Leuven. MM is supported by a KU Leuven Postdoctoral Mandate. MM and ND have been supported by an internal C1 grant of KU Leuven (C14:17/102), awarded to KA and BB. The work is further supported by FWO grants G0B0318N and G0C9521N awarded to BB, GN and EO.

CRediT authorship contribution statement

Ward Deferm: Writing – review & editing, Writing – original draft, Visualization, Software, Methodology, Investigation, Formal analysis, Data curation. Tiffany Tang: Writing – review & editing, Writing – original draft, Supervision, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Matthijs Moerkerke: Writing – review & editing, Supervision, Methodology, Data curation, Conceptualization. Nicky Daniels: Writing – review & editing, Project administration, Data curation, Conceptualization. Jean Steyaert: Writing – review & editing, Conceptualization. Kaat Alaerts: Writing – review & editing, Supervision, Methodology, Funding acquisition, Conceptualization. Els Ortibus: Writing – review & editing, Funding acquisition, Conceptualization. Gunnar Naulaers: Writing – review & editing, Funding acquisition, Conceptualization. Bart Boets: Writing – review & editing, Supervision, Resources, Methodology, Investigation, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: