This study was conducted with a prospective and cross-sectional design. To estimate the microstructural changes in the developing brain, we had a target age rating from 6 to 28 years. Participants with TSC were recruited from the Department of Pediatrics or Dermatology at Osaka University Hospital (Osaka, Japan) between September 2019 and January 2021.

All subjects met the diagnostic criteria for definite TSC. Subjects with a history of brain surgery (either epilepsy surgery or resection of subependymal giant cell astrocytomas) were excluded. Clinical information about epilepsy was obtained retrospectively from the medical records, including the following: history of ISs, epilepsy onset in months or the severity of epilepsy, the number of current anti- epileptic drugs (AEDs), and the use of mTOR inhibitors. Epilepsy type was classified using ILAE classification (28) as follows: developmental and epileptic encephalopathies (DEE) and focal epilepsy (FE), and the severity was classified as follows: 0, never; 1, well controlled (seizure-free for more than a year); and 2, refractory epilepsy. The cortical tubers were identified in fluid-attenuated inversion recovery (FLAIR) images. In addition, subjects with typical development who had no history of neurological conditions, including epilepsy or ASD, were recruited as control volunteers from neighboring cities. All participants and/or their guardians provided written informed consent to participate in the study. This study was approved by the Institutional Review Board of Osaka University Hospital (No. 18491) and was conducted according to the principles of the Declaration of Helsinki. All subjects were informed of the study and provided consent before study entry.

MRI images were acquired using a 3.0 T MRI scanner (SIGNA Architect; GE Healthcare, Milwaukee, WI, USA) for all subjects (n = 43). Single-shot, spin-echo, echo-planar imaging (EPI) was performed. Axial diffusion-weighted images were acquired with the following parameters: echo time (TE) = 75 ms, repetition time (TR) = 6,000 ms, matrix size = 128 × 128, field of view (FOV) = 260 × 260 mm², slice thickness = 3 mm, and number of slices = 50. Diffusion sensitization gradient was applied with 25 non-collinear gradient directions and a b value of 1,000 s/mm², in addition to one non-diffusion-weighted scan. The children who could not keep still for the scanning were sedated with either pentobarbital, thiopental, pentazocine, levomepromazine, risperidone, triclofos, diazepam, midazolam, or a combination of the afore mentioned sedatives.

The Mann-Whitney U test, chi-squared (χ²) test was used to compare the group descriptive data of the groups. Using randomize, which is an FSL tool for non-parametric permutation inference, voxel-wise statistical analysis was performed. Voxels were extracted where the FA value was significantly different between skeletons, using a general linear model (GLM) (41). Five-thousand permutations were applied, and the threshold was set to p < 0.05. The results were corrected for multiple comparisons using threshold-free cluster enhancement (TFCE) with family wise error (FWE) rate controlled with a cluster size of voxel > 10. Age and sex were used as covariates in the design matrix. Using the JHU White Matter Tractography Atlas, the anatomical regions of significant clusters were identified.

We hypothesized that the limbic area white matter would be associated with TAND, and we targeted seven regions based on our hypothesis: (1) the body of the fornix (FNX), (2) sagittal stratum (SS), (3) stria terminalis (ST), (4) cingulum in the cingulate gyrus (CGC), (5) superior longitudinal fasciculus (SLF), (6) uncinate fasciculus (UF), and (7) splenium of the corpus callosum (CC).

Correlations between DTI parameters (FA/MD) and behavior assessments (estimated IQ/socialization score /maladaptive behavior index/SRS-T score) were determined using Spearman's correlation coefficient (rs). Significance was established at p < 0.05. To reveal TAND related white matter in the TSC group (n = 19), multiple linear regression was performed to evaluate the association between independent variables (behavior) and dependent variables (FA/MD); age and sex were adjusted as covariates.

The model was as follows:

where y = FA/MD value, β₁, β₂, β₃= coefficient, β₀=intercept, Behavior = estimated IQ/socialization score/maladaptive behavior index/SRS-T score. In addition to above-mentioned 7 target regions, we examined the association between behavior and the anterior limb of internal capsule (ALIC), which connects prefrontal cortex to subcortical structures. To reveal active epilepsy influence on DTI parameters (FA/MD), we added refractory epilepsy presence (yes = 1, no = 0) as an independent variable in the above model. Age and sex were adjusted as covariates. All statistical analyses were performed using STATA/IC16.1 (Stata Corp LLC, College Station, Texas, USA).

We included 43 subjects in our analysis. A total of 19 subjects with TSC (mean age ± SD; 12 ± 6 years, range 6–28) were enrolled. Of the 27 subjects with TSC recruited to the TSC group, 8 were excluded (5 due to a history of brain surgery and, 3 due to MRI motion artifacts), leaving a final sample of 19 subjects in the TSC group. In addition, 24 age-matched control subjects (mean age ± SD; 12 ± 5 years, range 6–27) were recruited, and none were excluded (Figure 1). The TSC and control groups did not differ significantly in age or sex, but the estimated IQ was significantly lower in the TSC group (z = 4.4, p < 0.001; Table 1). The main clinical features of the 19 subjects with TSC are summarized in Table 2. Fourteen subjects (74%) had concomitant epilepsy, 6 and 11 of whom had DEE and FE, respectively. The seizure severity was classified to well controlled (seizure-free for more than a year) epilepsy in 9 subjects, and refractory epilepsy in 5 subjects. Refractory epilepsy frequency ranged from more than once a week to less than once per month. Among those with epilepsy, 6 (43%) had a history of ISs. All 19 subjects with TSC had cortical tubers on FLAIR images, 17 with multiple cortical tubers in both hemispheres and 2 with multiple cortical tubers in left hemisphere. FSIQ was unavailable for four subjects with TSC, because their intellectual functioning was below the floor of the Wechsler scales, although we estimated their IQ by using the Vineland Adaptive Behavior Score. Eight subjects (42%) had comorbid ASD of which 6 (aged 6–15 years) had an SRS-T score over 80, and two adult subjects (23 and 28 years old) exceeded the PARS-TR cut-off value. Chi-squared (χ²) test indicated that early- onset epilepsy (before 2 years) and a history of ISs were risk factors for ID (estimated IQ <70; p = 0.04 and, p = 0.03, respectively), but were not risk factors for ASD features (SRS-T score over 80; p = 0.12 and, p = 0.17, respectively). In addition, there was a positive correlation between the Maladaptive Behavior Index and the SRS-T score (rs = 0.76, p < 0.001). The type and rate of TAND behavioral level difficulties in subjects with TSC are summarized in Supplementary Material 1. Five subjects (26%) were treated with mTOR inhibitors at the time of MRI in the present study. The Mann-Whitney U test indicated that there was no significant difference between the TSC patients receiving mTOR inhibitors (n = 5) and those not receiving mTOR inhibitors (n = 14) in the maladaptive behavior index score (with mTOR inhibitor, mean ± SD: 19 ± 1.5; without mTOR inhibitor: 18 ± 2.1; z = −1.02, p = 0.31), SRS-T score (with mTOR inhibitor, mean ± SD: 73 ± 8.8; without mTOR inhibitor: 68 ± 19, z = −0.57, p = 0.57), estimated IQ (with mTOR inhibitor, mean ± SD: 56 ± 31; without mTOR inhibitor: 73 ± 32; z = 0.97, p = 0.33).

In the TSC brain, there are two main pathophysiological mechanisms for neurological manifestations: (1) focal abnormalities associated with cortical tubers, and (2) abnormal structural and functional connectivity.

Cortical tubers are lesions with cortical malformations, which result from the abnormal maturation and differentiation of neurons (44). These morphological changes originate from the overactivation of the mTOR pathway in the early stage of development. Dysmorphic neurons in the cortical tubers cause neurological dysfunction. Recently, fetal cerebral lesion scores have been reported to correlate with neurodevelopment and ASD in TSC (45). Therefore, TAND manifestations and microstructural disruption could result from cortical malformations. Cortical tubers can also induce epilepsy (46). Complete resection of the epileptogenic zone associated with cortical tubers has been successfully reported in TSC (47). IS, early-onset epilepsy, and refractory epilepsy were reported to be risk factors for TAND (12, 48, 49). In this study, IS and early-onset epilepsy were risk factors for ID but were not risk factors for ASD features. Studies suggested that severe or uncontrolled epilepsy cause poor cognitive outcomes (50, 51). Here, refractory epilepsy was associated with both FA and MD in the left ST, right SLF, right SS, FA in the left SS, SLF, MD in the bilateral CGC, and right UF. These findings coincide with a DTI study of children with TSC with persistent seizures (49). Our results suggest that refractory epilepsy influences the alternation of the white matter microstructure integrity in the frontal and temporal areas, which are strongly associated with intellectual ability and sociability. In contrast, the effect of epilepsy on ASD is unclear and requires further exploration (50–52). It is important to consider that the adverse effect of treatment of epilepsy affects the neuro-psychiatric manifestations in subjects with TSC (53). In this study, we did not compare the location or volume of cortical tubers with DTI parameters, because most of the patients showed multiple cortical tubers and had poorly defined borders. TSCs have been reported to exhibit abnormalities in structural and functional connectivity. Many animal studies have demonstrated the crucial role of the mTOR pathway, especially mTORC1, in oligodendrocyte differentiation (54, 55) and myelin gene expression (56, 57). A mouse model with oligodendrocyte-specific overactivation of mTORC1 showed decreased p4E-BP1, pAKT, and hypomyelination (56). These data suggest that mTORC1 activity disrupts myelination. A recent neuropathologic study reported a significant association between IQ at surgery and reduced numbers of oligodendroglia in the white matter of cortical tubers in subjects with TSC (58). Increased MD and decreased FA suggest abnormal myelination or disrupted microstructural organization of the white matter in subjects with TSC (3). In addition, two event-related potential (ERP) studies have reported atypical face processing in an individual with TSC (59, 60). The longer latency of ERPs indicated slower face processing, which suggested a disruption of myelination or weakness in neural connectivity.

The cumulative abnormalities in white matter microstructural integrity in the corpus callosum (CC) were associated with increasing neurological comorbidities, such as ID, epilepsy, and ASD in subjects with TSC (15), and the comorbid diagnosis of ASD was associated with the largest change in FA (15). Several studies have reported that there was a significant difference in the microstructural integrity of the CC between children with idiopathic ASD and healthy controls, and the DTI parameters of CC were correlated with the socio-communicative ability (61, 62). In addition, significant differences in FA in the arcuate fasciculus between TSC subjects with and without ASD have been reported (63). A longitudinal study of infants with TSC (0–2 years old) revealed that subjects with TSC who develop ASD exhibit reduced FA in several major white matter tracts, including the arcuate fasciculus, CC, and the SS, relative to TSC subjects without ASD (64). These findings suggest abnormal myelination or disrupted microstructural organization of white matter in the developing brain. This study showed significantly reduced FA in the right association fibers such as the IFOF, ILF, SLF, UF, and CC in the TSC group compared with the healthy control group. The IFOF connects the occipital lobe and OFC. The short fibers of the ILF connect the occipital lobe and temporal lobe, and the long fibers of the ILF connect the visual areas of the occipital lobe to the amygdala and hippocampus (65). The IFOF and ILF have been repeatedly reported to be critically important for face processing (66). The SLF connects the parietal, occipital, and temporal lobes (66), and it is related to the major cognitive abilities, such as language, attention, and memory; in particular, the right SLF is associated with visual perception (67, 68). Furthermore, the OFC is important for the recognition of expressed emotions (69). The UF connects the temporo-amygdala-orbitofrontal network (22); in particular, the right UF is strongly associated with the emotional empathy network (23). Therefore, we hypothesize that reduced FA in the right association fibers in subjects with TSC indicates socio-emotional difficulties such as a lack of emotional empathy, poor face processing, or unrecognition of expressed emotion. These behavioral features may be considered as the parts of TAND. On the other hand, FA in the pontine crossing tract and bilateral CST were significantly higher in the TSC group than in the control group. A limitation of DTI is the difficulty in representing crossing white matter fascicles (70). Generally, crossing fiber problems occur when there are more than two fiber bundles that are differently oriented in the same voxel (71), such as SLF and CST, which are crossing in centrum semiovale (72, 73). FA is sensitive to microstructural change (74), and FA decrease reflects the reduced white matter integrity. However, where nerves cross, an atypical increase in FA may occur owing to a higher apparent diffusion coefficient of the other nerve when one nerve is damaged (75). For example, DTI studies of subjects with both multiple sclerosis (MS) and Alzheimer's diseases reported a higher FA than healthy controls in the crossing fiber regions when one of the fiber bundles was degenerated (75, 76).

Here, TBSS showed a higher FA in the bilateral CST and pontine crossing tract, which are the crossing fiber regions. We speculate that one damaged fiber bundle in the crossing fiber regions may induce an atypical increase in FA in subjects with TSC.

Bolton et al. (77) reported that developmental abnormalities in the temporal lobe, resulting from temporal tubers could be a risk factor for comorbidity with ASD, because of impairments in visual recognition, such as facial expression. “The amygdala theory of autism” proposed that abnormalities in the amygdala could be a cause of autism, which suggested that three regions, 1 the amygdala, 2 the OFC, and 3 the superior temporal sulcus and gyrus, play an important role in social intelligence (78, 79). The TAND psychiatric level includes ASD, ADHD, anxiety disorders, and depressive disorders (9). In this study, the SRS-T score, which represents autistic features, was associated with FA and MD in the FNX, FA in the bilateral ST and right ALIC, and MD in the splenium of the CC. A previous mega analysis DTI study of psychiatric disorders revealed that there was a common white matter alternation in subjects with schizophrenia and bipolar disorder in the limbic system, including the fornix and cingulum (80). The hippocampus and the mammillary body are connected by the body of the fornix, and the mammillary body and anterior nuclei are connected by the mammillothalamic tract, which forms the hippocampal-diencephalic-cingulate network associated with memory, emotion, and psychiatry (21, 22). The ST is a small limbic pathway that forms the main efferent pathways from the amygdala, and run parallel to the fornix (21). The ST combines the amygdala with the bed nucleus of stria terminalis (BNST) (81). The BNST is the hub for integration of limbic information and contains many sub-nuclei. The BNST is important for detection and expression of fear and, regulates emotional state or arousal (82). All the prefrontal cortex fibers (both ascending and descending) are projecting to the ALIC, which is one of the most established targets of deep brain stimulation (DBS) or other neuro-surgical treatment for neuropsychiatric disorders (26, 83). The dysfunction of the cortico-striato-thalamo-cortical (CSTC) loop is common model for neuropsychiatric disorders (84). Recently, the effects of DBS to the ALIC, the nucleus accumbens (NAcc), basolateral amygdala, and hypothalamus for self-injury behavior in subjects with ASD were reported (85, 86), and these areas are in anatomical proximity with the targets of DBS for obsessive compulsive disorder(OCD) (87). In this study, TAND behavioral level difficulty was associated with FNX and ST, and the ALIC was associated with autistic features. These results suggest the existence of aberrant neural network, which may be relevant to behavioral problems, such as anxiety, severe aggression, or self-injury, in subjects with TSC, as well as other neuropsychiatric disorders.

Recently, mTOR-inhibitors, that block mTORC-1 have been approved for the manifestations of TSC, such as renal angiomyolipoma, SEGAs, and epilepsy (4, 88, 89). Although mTOR-inhibitors show definite efficacy in epilepsy, their effectiveness in TAND remains controversial (88, 90, 91). In a human study, primary mixed glial cell cultures derived from patients with TSC or focal cortical dysplasia, which is a histologically close condition, reduced oligodendroglial turnover in association with a lower myelin content. In addition, mTOR inhibitors repair decreased myelination (92). Two studies have reported that mTOR inhibitors change the white matter microstructure (88, 89). The efficacy of everolimus for intractable epilepsy and the improvement of autistic features in patients with TSC have been reported in a previous Japanese study (90). In this study, five subjects with TSC (26%) were treated with an mTOR-inhibitor (four had refractory epilepsy, one was the treatment of SEGA). Although we considered the effect of mTOR inhibitors on white matter integrity, we included these patients because this study focused on the association between TAND manifestations and white matter microstructure regardless of treatment. As a result, the FA value of major white matter tracts was not high in the subjects using mTOR inhibitor (Supplementary Material 8). To evaluate the effects of mTOR-inhibitors on white matter microstructure, we should observe the change in DTI parameters before and after treatment with mTOR-inhibitors. In the future, the effect of mTOR-inhibitors on white matter microstructure associated with TAND manifestations should be examined in longitudinal studies with a large cohort. Particularly, investigating whether the change in behavior score coincides with changes in the limbic white matter and neural networks may reveal the effects of mTOR-inhibitors on TAND.

The present study has several limitations. First, our sample size was small, and the study design was not longitudinal. When we consider the TAND as a neural network disorder, TBSS is not adequate to identify specific neural network to basal ganglia and/ or brainstem, which recently got attention for the association with neuropsychiatric disorders. The future studies with a large cohort and advanced neuro-imaging techniques are required. Longitudinal follow-up is needed to confirm and extend our results. Second, the effects of genotype were not considered. Since we did not obtain genetic information, we could not consider any genotype-phenotype correlations. As a previous study reported genotype-TAND correlations in TSC2 (9), the correlation of genotype should be performed in the future. Third, we did not consider the effect of tubers on DTI analysis. In the future, positron emission tomography should be studied to reveal the effect of tubers on TAND (81).