Patients with unilateral TLE were enrolled from the Department of Neurology at the First Affiliated Hospital of Guangxi Medical University between January 2017 and September 2021. According to the classification standard guidelines of 1981, 1989 and 2017 formulated by ILAE (26), the secondary generalization diagnosis of complex partial seizures and partial seizures was performed. Inclusion criteria were as follows: (1) all patients had unilateral (left and right) TLE, which was confirmed by MRI structural image, video electroencephalography (EEG) assessment and clinical manifestation analysis. (2) All patients took antiepileptic drugs (AEDs) regularly; (3) All patients were right-handed. Exclusion criteria were as follows: (1) comorbidities affecting cognitive function, including traumatic brain injury, intracranial tumor, stroke, infection, multiple sclerosis, and Alzheimer's disease; (2) Patients with a score <24 on MMSE; (3) patients who take or are taking topiramate and barbiturates; (4) MRI structural images showing abnormalities except hippocampal sclerosis.

Every patient had FIAS and/or FAS, with some also having FBTCS. Patients were divided into two groups for this study based on their medical history at the time of scanning. The “non-FBTCS” group included 25 patients who had never experienced FBTCS, the “FBTCS” group included 25 patients with recurrent FBTCS in the year before scanning, and 25 healthy control subjects with matching demographic features were recruited as a neuroimaging reference group. As confirmed by health screening techniques, the control group had no psychological or neurological issues. All procedures were approved by the Ethics Committee of Guangxi Medical University's First Affiliated Hospital. All individuals provided written informed consent for the study.

The Montreal Cognitive Assessment (MoCA) was used to evaluate cognitive impairment.

The ANT was used to assess each subject's vigilance, as previously stated (27). Participants were instructed to keep their eyes on a fixation cross in the screen's center and determine the direction of the target arrow throughout the trials. Participants were told to press a button to provide an answer as accurately and quickly as they could. The formal test included three blocks of 96 trials, plus the practice block. The entire test took approximately 25 min. In the test, participants had to decide whether the middle arrow would point left or right next. They were given three types of hints: (1) a center cue, characterized by the presence of an asterisk at the central fixation point; (2) double cues, characterized by an asterisk positioned above and below the fixation cross; or (3) no cue, characterized by a pair of arrows pointing in the opposite direction as the target arrow, or a pair of dashes, flanking the fixation point. Each test contained unique clues, targets, and surrounding interference data, and they were presented in a random order. The device automatically detected and recorded participants' reaction time (RT). the no-cue condition indicated tonic alertness and represented a state of general wakefulness, similar to sustained attention. The double-cue condition indicated phasic alertness and represented the ability of the participant to be response ready for a short period of time subsequent to the presentation of external cues or stimuli. Alertness was reflected by the RT in the two different warning conditions. Alertness was determined by subtracting the median of the double cue condition from the median of the no cue condition. The larger the alertness value was, the greater the degree of alertness.

The images of the participants were acquired at the First Affiliated Hospital of Guangxi Medical University utilizing a 3.0 Tesla MRI scanner (Philips, The Netherlands) equipped with a standard eight-channel head coil. Throughout the scan, all individuals were instructed to close their eyes and relax but not to sleep. Foam cushions and noise-canceling earplugs were used to reduce noise and head movements. A T1-3D BRAVO sequence was used to acquire high-resolution sagittal T1WI images with the following acquisition parameters: repetition time (TR) = 7.8 ms, echo time (TE) = 3.4 ms, flip angle = 9°, field of view (FOV) = 256 × 256 mm, matrix = 256 × 256 mm, slice thickness = 1 mm without slice gap, voxel size = 0.89 × 0.89 × 1 mm, and 176 sagittal slices. Resting-state functional MRI images were collected by using gradient-echo single-shot echo-planar imaging sequences with a TR = 2,000 ms, TE = 30 ms, FOV = 220 × 220 mm, FA = 90°, matrix = 64 × 64, slice thickness = 3.5 mm, slice gap = 0.5 mm, voxel size = 3.44 × 3.44 × 4 mm, 41 slices, and 225 volumes. For data quality control, the scan was evaluated by two professional neuroradiologists who were blinded to the clinical information.

SPSS 20.0 (SPSS Inc., Chicago, IL, USA) was used to perform statistical analyses. To compare normally distributed data among the three groups (P < 0.05), one-way analysis of variance (ANOVA) was employed, and chi-square tests were utilized to analyze categorical variables. To compare clinical factors between the two groups of patients, Student's t-tests for normally distributed variables and Mann–Whitney tests for non-normally distributed data were used. P < 0.05 was chosen as the level of statistical significance.

We used DPABI's statistical analysis toolkits to compare FC and GMV maps for each ROI among the three groups. ANOVA was used to compare the imaging variables among the groups, with age, sex, and head motion as covariates (P < 0.05). Multiple comparisons were corrected using a false discovery rate (FDR) correction for clusters with more than 30 voxels. Then, pairwise comparisons of regions with significant group differences in FC were undertaken. We used two-tailed paired comparison t-tests and a false discovery rate (FDR) correction at P < 0.05. Correlation analyses between imaging characteristics and clinical factors for patients were conducted in SPSS 20.0, revealing substantial group differences. Multiple comparisons were adjusted using the Bonferroni method. Parametric comparisons were performed with Pearson's correlation analysis, and non-parametric comparisons were performed with Spearman's correlation analysis.

To ascertain the ability of seed-based rsFC to discriminate among the three groups, multiclass discriminant analysis was performed using PRoNTo v2.0 software in MATLAB 2018b (28). Specifically, the multiclass classification was transformed into three binary classifiers using a one-vs.-one coding method. To reduce the dimensionality of initial features, starting features were selected from voxels with significant group effects (P < 0.05, uncorrected). After that, the outputs of all binary classifiers were combined using an error-correcting output code technique. A 10-fold cross-validation approach was used to assure generalization during this process. Finally, we calculated the total and group-specific accuracies.

The demographic and clinical data of all participants are summarized in Table 1. No differences in age, sex, handedness, seizure focus, age of seizure onset, disease duration, or seizure frequency were discovered among the three groups (P > 0.05). However, the RTs on the no cue and double cue conditions were significantly different among these three groups. A post-hoc test revealed no significant differences between FBTCS and non-FBTCS patients but significant differences between FBTCS or non-FBTCS patients and HCs. In addition, the RTs of the TLE group were longer than those of the HC group.

There was no significant difference in GMV among the three groups for the BF subregions (P > 0.05 [FDR corrected]).

All four seeds exhibited significant group effects (Figure 1 and Table 2). Specifically, significant alterations in rsFC were detected in four clusters in the left Ch1-3, including rsFC with the right cerebellar posterior lobe, bilateral striatum, left superior frontal gyrus, and right middle temporal gyrus. Significant alterations in rsFC were found in four clusters of the right Ch1-3, primarily rsFC with the right cerebellar posterior lobe, bilateral striatum, right precuneus, and left middle occipital gyrus. The bilateral amygdala had significant anomalies in rsFC with the left Ch4, and the left amygdala had significant anomalies in rsFC with the right Ch4.

With the left Ch1-3 as a seed, the FBTCS and non-FBTCS patients showed considerably lower FC than HCs with the right cerebellum posterior lobe, bilateral striatum, left superior frontal gyrus, and right middle temporal gyrus, respectively, and the FBTCS group demonstrated decreased FC compared to HCs with the right cerebellum posterior lobe, bilateral striatum, and left superior frontal gyrus, respectively (Figure 2A). Using the right Ch1-3 as a seed, the FBTCS and non-FBTCS groups showed significantly decreased FC compared to HCs with the right cerebellar posterior lobe, bilateral striatum, right precuneus, and left middle occipital gyrus, respectively, and the FBTCS group showed lower FC than the non-FBTCS group with the bilateral striatum (Figure 2B). When the left Ch4 was used as a seed, the FBTCS group showed significantly lower FC than the non-FBTCS and HC groups with the bilateral amygdala (Figure 2C). Using the right Ch4 as a seed, the FBTCS group showed significantly decreased FC with the right amygdala compared with the non-FBTCS and HC groups (Figure 2D).

To further explore whether there were differences in the brain regions between left and right mTLE in the different TLE groups, an ANOVA was performed. We found that there was no significant difference (P > 0.05) between the two TLE groups. Details are shown in Supplementary Material 1.

The rsFC between the left Ch1-3 and striatum had a significant positive correlation with performance in the double cue (r = 0.48, P = 0.015) and no cue conditions (r = 0.495, p = 0.012) in the FBTCS group. In the non-FBTCS group, the rsFC between the left Ch4 and striatum had a moderate negative correlation with performance in the double cue (r = −0.458, P = 0.021) and no cue conditions (r = −0.507, p = 0.0097); additionally, the rsFC between the right Ch1-3 and striatum had a moderate negative correlation with performance in the no cue condition (r = −0.44, p = 0.028) (Figure 3).

Table 3 summarizes the overall and group-level accuracy. In general, among fourth basal forebrain subregions to voxel functional connectivity in differentiating subjects from each other, the right Ch1-3 and left Ch4 subregions performed better than the left Ch1-3 and right Ch4 subregions in identifying the three groups (accuracy 85.33 and 81.00% vs. 78.67 and 78.67%).

The BF is located in the front of the forebrain, beneath the striatum. The BF has many cholinergic projections to the neocortex, which is involved in the neuromodulation of cognitive performance. In this study, there was no difference in the volume of the BF subregions among the three groups. Memory loss has been linked to Ch4 neuron deterioration in neurodegenerative disorders such as Alzheimer's disease, mild cognitive impairment, Parkinson's disease with moderate cognitive impairment, and Wilson disease (29, 30). However, Ma et al. found no difference in the volume of BF subregions in a previous study on short-term and chronic insomnia (17). There has been no research on the volume of the basal forebrain in TLE until now. We found no differences in BF volume among the three groups. This could be because TLE originates mostly in the hippocampus and amygdala and has less direct influence on the BF or because differences in volume take a longer time to manifest. The small sample size of this study is another possible explanation for our findings. More information about changes in basal forebrain volume could be achieved from further studies with larger sample sizes and longitudinal designs.

The BF nuclei, which contain four distinct cell groups (Ch1–4), are the primary sources of cholinergic projections to the neocortex, amygdala, and hippocampus (31). In comparison to the non-FBTCS group, we discovered that FC was reduced between Ch1-3 and the bilateral striatum as well as the left posterior cerebellar lobe (PCL) in FBTCS patients; additionally, FC was reduced between Ch4 and the amygdala. The striatum is crucial for a number of complex functions, ranging from motor control to action selection and attention (32). Electron microscopy demonstrated that the ventral and dorsal striatum provide synaptic input to cholinergic BF neurons (33). A neuropathological analysis established a direct anatomical relationship between the striatum and basal forebrain, providing behavioral and structural evidence (34). Striatal structure and function are altered in attention-deficit/hyperactivity disorder (35) and a variety of epileptic conditions, including focal to bilateral tonic-clonic seizures (36) and pediatric epilepsy (37). Additionally, we found a significant correlation between the rsFC of Ch1-3 and the striatum and vigilance measurements. We postulate that recurrent seizures have a detrimental effect on the striatum, reducing ACh levels in the basal forebrain and impairing vigilance. Numerous studies have demonstrated that the cerebellum is essential for normal cognitive function. A lesion in the PCL can result in cerebellar cognitive-affective syndrome, characterized by issues with executive function, visual-spatial processing, linguistic abilities, and emotional regulation. In TLE with or without FBTCS as well as right TLE, we previously described aberrant FC between deep cerebellar nuclei and the cerebral cortex (38, 39). In the current study, we found decreased rsFC between Ch1-3 and the left posterior cerebellar lobe in FBTCS patients, which is consistent with earlier findings. Our findings imply that disruption of the rsFC between the basal forebrain and cerebellum may contribute to cognitive impairment in TLE. Prior research has established that the amygdala regulates the prefrontal cortex (PFC) and hippocampus directly via cholinergic projections from the basal forebrain and subsequent acetylcholine release (40). According to Adam et al. (40), cholinergic projections from the basal forebrain moderate activity in the greater amygdala during the processing of physiologically relevant stimuli in humans. The Ch4 subregion has cholinergic projections to the neocortex and amygdala. In our investigation, we found that the rsFC between Ch 4 and the amygdala was considerably lower in the FBTCS group than in the non-FBTCS group. We postulate that repeated FBTCS seizures disturb the basal forebrain-amygdala circuit by reducing the amount of ACh projected from the basal forebrain to the amygdala. Together, aberrant rsFC between basal subregions and the striatum, cerebellum posterior lobe, and amygdala influence alertness and may be a critical neuroimaging biomarker for differentiating FBTCS and non-FBTCS patients.

The DMN is involved in working memory, emotions, cognitive performance, and epileptic activity. Mounting data suggest that disrupting the DMN may cause epileptic activity, cognitive dysfunction, and mental dysfunction in TLE patients (41). The precuneus and middle temporal gyrus (MTG) are key components of the DMN, as they participate in processes related to consciousness, self-reflection, visuospatial function, and cognition (42–44). Nair et al. (45) demonstrated that gamma oscillations were restricted to the BF and that BF gamma-band activity had a direct effect on a DMN hub in rats, implying that the BF may be an important target for DMN regulation. In our study, both groups of patients displayed significantly lower rsFC between the right Ch1-3 and right precuneus as well as left MTG in comparison to HCs, indicating decreased synchronous neuronal activity between the DMN and basal forebrain. We speculate that the DMN might be a target for epilepsy and cognitive control in the basal forebrain. Both the DMN and the basal forebrain subregions influence cognitive function, suggesting that disrupted rsFC from the basal forebrain to the DMN could impact cognition in TLE patients.

In addition, compared to HCs, both patient groups had lower rsFC between Ch1-3 and cortical regions, such as the left superior frontal gyrus (SFG) and left occipital middle gyrus, indicating reduced cholinergic innervation in these cortical regions. Damage to the SFG, which is a crucial component of the frontoparietal network, might cause vigilance deficits. The FC between Ch1-3 and the left SFG was considerably lower in the patient groups than in the HCs, implying that diminished ACh in the frontal cortex may contribute to cognitive impairment in TLE. The findings were in line with earlier research that showed disturbed rsFC in the frontal cortices, linking it to deficits in alertness (46) and executive function (47) in TLE. fMRI studies have demonstrated that the visual cortex is engaged and cerebral blood flow is elevated during attention-related tasks (48, 49). In rTLE patients, the functional activity of the superior occipital gyrus in the alertness-related network was higher than that in HCs (46). Julia Schumacher et al. also identified aberrant functional connectivity between the basal forebrain and occipital cortex in Lewy body dementia and Alzheimer's disease, which they believe may indicate a cholinergic system imbalance and a shift in the cholinergic input to the occipital cortex (22). Additionally, gray matter volume reduction and hypometabolism were detected in the visual cortex of patients with TLE (50, 51). Therefore, our findings support the concept that in TLE the basal forebrain improperly modulates the frontoparietal and sensory networks, resulting in cognitive dysfunction.

This investigation has some limitations. First, the sample size is relatively small. As a result, our findings need to be confirmed in a broader patient group. Second, this was a cross-sectional study. Therefore, a longitudinal assessment of resting-state fMRI in temporal lobe epilepsy will be required to confirm these findings. Third, we did not take into account the effect of antiepileptic medicines on FC and VBM. Fourth, we cannot rule out the effect of interictal discharges on patient alertness because a synchronous electroencephalogram was not performed during the acquisition of imaging data. Finally, we defined BF subregions using a probabilistic map extracted from the SPM Anatomy Toolbox, which has been extensively used in previous research (1, 17). However, the BF subregions are segregated in additional ways (52). In the future, it is vital to employ different parcellation methods to acquire a thorough understanding of BF alterations in TLE.