Disorganization Of Language And Working Memory Systems in Frontal Versus Temporal Lobe Epilepsy Part 1

Cognitive impairment is a common comorbidity of epilepsy and adversely impacts people with both frontal lobe (FLE) and temporal lobe (TLE) epilepsy. While its neural substrates have been investigated extensively in TLE, functional imaging studies in FLE are scarce. In this study, we profiled the neural processes underlying cognitive impairment in FLE and directly compared FLE and TLE to establish commonalities and differences. We investigated 172 adult participants (56 with FLE, 64 with TLE and 52 controls) using neuropsychological tests and four functional MRI tasks probing expressive language (verbal fluency, verb generation) and working memory (verbal and visuo-spatial). 

Epilepsy is a neurological disorder that affects a person's physical and cognitive functions. Many people think that epilepsy affects memory, but this is not entirely true. Although people with epilepsy may experience some short-term memory problems, most patients' long-term memory is not significantly affected. There are even some studies that suggest epilepsy may improve memory in some cases.

People with epilepsy can regain their memory and cognitive function after treatment. Treatment for epilepsy often includes medications and surgery, which can help control symptoms and reduce the impact on a patient's memory and other cognitive functions. After receiving treatment, many patients find that their memory and other cognitive abilities improve, making it easier for them to make good decisions in life.

Additionally, some research suggests that epilepsy may promote improved memory. Research has found that in some cases, people with epilepsy experience miraculous memory effects known as "paranormal memory." These effects may improve patients' memory, making it easier for them to remember things and make better decisions.

Therefore, epilepsy is not a completely negative disease and it does not necessarily affect the patient's memory and cognitive function. In fact, after treatment, many patients can regain their memory and cognitive abilities, and in some cases improve memory. If you or a loved one has epilepsy, make sure you receive prompt medical treatment to relieve symptoms and improve your quality of life. It can be seen that we need to improve our memory. Cistanche deserticola can significantly improve memory because Cistanche deserticola is a traditional Chinese medicinal material with many unique effects, one of which is to improve memory. The efficacy of minced meat comes from the various active ingredients it contains, including acid, polysaccharides, flavonoids, etc. These ingredients can promote brain health in various ways.

Click Know to improve short-term memory

Patient groups were comparable in disease duration and anti-seizure medication load. We devised a multiscale approach to map brain activation and deactivation during cognition and track reorganization in FLE and TLE. Voxel-based analyses were complemented with profiling of task effects across established motifs of functional brain organization: (i) canonical resting-state functional systems; and (ii) the principal functional connectivity gradient, which encodes a continuous transition of regional connectivity profiles, anchoring lower-level sensory and transmodal brain areas at the opposite ends of a spectrum. We show that cognitive impairment in FLE is associated with reduced activation across attentional and executive systems, as well as reduced deactivation of the default mode system, indicative of a large-scale disorganization of task-related recruitment.
The imaging signatures of dysfunction in FLE are broadly similar to those in TLE, but some patterns are syndrome-specific: altered default-mode deactivation is more prominent in FLE, while impaired recruitment of posterior language areas during a task with semantic demands is more marked in TLE. Functional abnormalities in FLE and TLE appear overall modulated by disease load. On balance, our study elucidates neural processes underlying language and working memory impairment in FLE, identifies shared and syndrome-specific alterations in the two most common focal epilepsies and sheds light on system behaviour that may be amenable to future remediation strategies.
1 Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA 

2 Department of Clinical and Experimental Epilepsy, UCL Queen Square Institute of Neurology, London WC1N 3BG, UK 

3 MRI Unit, Epilepsy Society, Chalfont St Peter, Buckinghamshire SL9 0RJ, UK 

4 Multimodal Imaging and Connectome Analysis Laboratory, McConnell Brain Imaging Centre, Montreal Neurological Institute, Montreal, Quebec H3A 2B4, Canada 

5 Department of Neurology, Ludwig-Maximilians-Universität, 81377 Munich, Germany 

6 Epilepsy Unit, Hospital Clínic de Barcelona, IDIBAPS, 08036 Barcelona, Spain 

7 Department of Psychiatry and Psychotherapy, Central Institute of Mental Health, Medical Faculty Mannheim, University of Heidelberg, Mannheim, Germany 

8 Department of Neurology, Medical University of Vienna, Vienna, Austria 

9 Centre for Medical Image Computing, University College London, London, UK 

10 Neuroradiological Academic Unit, UCL Queen Square Institute of Neurology, University College London, London, UK 

11 Department of Medicine, Division of Neurology, Queen’s University, Kingston, Ontario, Canada 

12 Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA 

13 Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, PA 19104, USA 

14 Department of Neurology, University of Pennsylvania, Philadelphia, PA 19104, USA 15 Department of Psychiatry, University of Pennsylvania, Philadelphia, PA 19104, USA 16 Santa Fe Institute, Santa Fe, NM 87501, USA

Introduction

Frontal lobe epilepsy (FLE), the second most common focal epilepsy syndrome after temporal lobe epilepsy (TLE), is frequently drug-resistant and MRI-negative.1–3 Cognitive impairment is common in both FLE and TLE and adversely impacts quality of life and psychosocial functioning.4 Impaired episodic memory and semantic knowledge are common in TLE, although dysexecutive traits frequently coexist.5,6 In contrast, FLE has a less established cognitive signature, with multiple cognitive domains being affected, including dexterity, attention, working memory, verbal fluency, executive functions and episodic memory.7–11 Whether cognitive profiles in FLE and TLE may be distinct remains controversial, and several investigations concluded that these syndromes cannot be discriminated against based on cognitive measures.8,12,13 It is however suggested that episodic memory impairment is more profound in TLE, while executive functions may be more affected in FLE.7,14–16
Task-based functional MRI (fMRI) probes the neural correlates of cognitive impairment in epilepsy. In TLE, altered parietal and mesiotemporal activation and connectivity underlie working memory impairment,17,18 while language fMRI studies indicate altered frontotemporal functional profiles, with complex intra- and interhemispheric reorganization.19–25 In contrast, few studies investigated FLE.26 Children with FLE had reduced fronto-temporo-parietal connectivity during working memory fMRI but no substantial alteration in regional activation.27 In adults with FLE, we previously reported enhanced frontotemporal activation during episodic memory encoding and reduced mesiotemporal activation in those with poorer memory.28 

Abnormal motor and parietal activity may underlie impaired dexterity.29 Overall, a comprehensive overview of the neural substrates of cognitive impairment in FLE is lacking.

Here, we aimed to characterize the functional neuroanatomy of expressive language and working memory, cognitive functions reliant on frontal lobe processing,30,31 in individuals with drug-resistant FLE who underwent neuropsychological tests and four fMRI tasks. We compared people with FLE to (i) healthy controls, and (ii) a ‘patient control group’ of individuals with TLE, comparable in epilepsy duration and anti-seizure medication (ASM) load, which allowed us to establish shared and syndrome-specific traits. 

We devised a multiscale functional mapping framework to investigate the landscape of brain activation and deactivation during cognition 32–34 and capture disease-related reorganization. Under an ensemble view on the reconfiguration of task-related brain activity, we complemented traditional voxel-based fMRI maps, which elucidate task-related signatures at a regional level, by profiling task effects across two motifs of brain organization: (i) established resting-state functional systems35; and (ii) the principal functional connectivity gradient.36,37 

The gradient, in particular, describes a continuous transition of neural function that anchors unimodal sensory areas and high-order transmodal regions at two opposite ends of a spectrum, providing an axis of subregional cortical organization and recapitulating established models of cortical hierarchy.38 

Thus, the gradient offers a compact, yet formal, framework to characterize organizational aspects of cognitive activity, which allows us to (i) describe task-fMRI signatures in the context of a global balance of sensorimotor and highorder, perceptually-decoupled processing, as exemplified by work in healthy adults39–41 and people with TLE performing a pattern separation task42; and (ii) derive global metrics that quantify group differences in task-related systems-level reorganization. By conveying regional, systems-level and global viewpoints on the neural signatures of cognitive impairment in epilepsy, our approach collectively proves sensitive to both localized and higher-order abnormalities.

We anticipated expressive language and working memory impairment in FLE. We hypothesized that such impairment would be underpinned by (i) reduced activation of areas engaged during task execution, i.e. ‘task-positive’ regions; (ii) reduced deactivation of default-mode areas (DMN), i.e. ‘task-negative’ regions; and (iii) global disorganization of cognitive system recruitment, as quantified via gradient analyses. We also hypothesized that, based on the proximity to the epileptic focus, (i) frontal and systems-level working memory abnormalities may be more prominent in FLE than TLE; (ii) language-related activation of frontal areas would be lower in FLE; and (iii) engagement of temporal language areas would be lower in TLE. 

We also aimed to corroborate the neurobehavioural validity of our fMRI tasks by correlating imaging patterns with neuropsychological and task performance measures. Finally, we explored associations between cognitive network alterations and clinical characteristics, probed the potential effects of frontal lobe lesions, and replicated our main FLE findings in a more homogeneous patient subgroup with frontal cortical dysplasia.

Materials and methods

This study investigated 172 participants recruited from 2007 to 2013: 120 drug-resistant patients under surgical consideration, 56 with FLE (29 female, 30/26 left-/right-sided FLE), 64 with TLE (44 female, 34/30 left-/right-sided TLE) and 52 healthy controls (30 female) without neurological or psychiatric diagnoses and no family history of epilepsy. Demographic and clinical details are provided in Table 1.

Diagnosis of FLE was determined by expert epileptologists based on history, seizure semiology, video-EEG telemetry and 3 T structural MRI; PET, ictal single-photon emission computerized tomography (SPECT) and magneto-encephalography data were available for a patient subset. In 29 patients, MRI was nonlesional (left/right: 17/12). Findings in the remainder of patients included areas of suspected focal cortical dysplasia (FCD, n = 13; left/right: 6/7; pathologically confirmed in 8 of 8 patients who subsequently had surgery); dysembryoplastic neuroepithelial tumour (DNET, n = 6; left/right: 3/3); low-grade glial tumour (n = 3, all right); possible periventricular nodular heterotopia (n = 1, left); or unequivocal signal abnormalities, concordant with clinical and EEG findings [n = 4, left/right: 3/1; one post-traumatic, one of intrauterine (vascular) aetiology and two areas of cortical injury of unclear aetiology]. A lesion frequency map43 is shown in Fig. 1.

In people with TLE, interictal and ictal scalp video-EEG confirmed and lateralized seizure onset to the temporal lobe. All had ipsilateral hippocampal sclerosis on 3 T MRI, as determined by qualitative neuroradiological diagnosis and/or via quantitative assessments of hippocampal volumes44 and T2 relaxation times,45 with pathological confirmation in those who subsequently underwent surgery. Hippocampal sclerosis coexisted with ipsilateral DNET in three patients (left/right: 2/1) and a possible FCD in one patient (right).

Written informed consent was obtained from all participants according to the standards of the Declaration of Helsinki. Participant recruitment was approved by the University College London Queen Square Institute of Neurology and the University College London Hospitals Research Ethics Committee. Exclusion criteria were non-proficiency in written and spoken English, MRI contraindications, pregnancy and inability to give informed consent. Individuals who experienced focal to bilateral tonic-clonic seizures (FBTCS) <24 h before the investigation were excluded or had their testing session rescheduled.
Groups were comparable for handedness and (binary) sex, but not for age, which was used as a covariate in all group analyses. Patient groups did not differ in age at seizure onset and epilepsy duration, number of ASMs and usage of levetiracetam or topiramate/zonisamide, which more favourably or unfavourably influence cognitive system activity than other common ASMs, respectively.46,47 Patients with FLE had more frequent seizures, shorter time since the last seizure and a more frequent history of FBTCS in the year before the investigation than those with TLE (Table 1). As FLE is heterogeneous in terms of aetiology and MRI findings, we separately analysed a subgroup with a more homogeneous aetiology (FCD; n=13), directly compared FLE patients with and without lesions and probed the influence of clinical variables on imaging findings. Moreover, we separately investigated left and right FLE subgroups.

Neuropsychological data

Participants underwent standardized neuropsychological tests,48 providing measures of general intellectual level (IQ, National Adult Reading Test49), working memory [digit span and Wechsler Adult Intelligence Scale (WAIS-III)50 scores], letter and category fluency51 (sum of words generated for letter ‘S’, sum of items generated for the category ‘Animals’ in 1 min), naming (McKenna Graded Naming Test52), psychomotor speed and executive function (mental flexibility; Trail Making Test A and B-A53) and verbal and visuospatial learning and recall (List and Design Learning, A1–A5 and A6, Adult Memory and Information Processing Battery54). Verbal reasoning and comprehension measures (Vocabulary and Similarities, WAIS III50 scaled scores) were available for patient groups. Pairwise deletion was used for missing data.

Imaging data acquisition and fMRI tasks

Imaging data were acquired on the same GE SignaHDx 3T MRI scanner at the Epilepsy Society, Chalfont St Peter, Buckinghamshire, UK. For all tasks, we used a 50-slice gradient echo-planar sequence with axial orientation, 64×64 matrix, in-plane voxel size 3.75×3.75 mm, 2.4 mm slice thickness, 0.1 mm inter-slice gap, echo time/repetition time: 25/2500 ms.55 One visuospatial and one verbal fMRI paradigm assessed working memory. During the visuospatial (Dot Back) task, dots appeared in four possible locations on a screen. Participants were instructed to move a joystick to the position of the currently presented dot (0 Back) or the position of the dot displayed one (1 Back) or two presentations earlier (2 Back).55 There were five 30-s blocks for each condition in pseudo-random order, intermixed with 15 s of cross-hair fixation. 

During the verbal working memory task, single concrete nouns were displayed every 3 s within 30 s blocks. Participants responded upon display of a given control word (active control condition) or the recurrence of a word displayed two presentations earlier (2 Back working memory). There were five 30 s blocks per condition, intermixed with 15 s of cross-hair fixation. Two covert (silent)56 tasks probed expressive language, followed by out-of-scanner cognitive testing in the same session. During verbal fluency fMRI, participants generated words beginning with a visually-presented letter (A/D/E/S/W, one letter per block, five 30 s blocks), alternating with 30 s blocks of cross-hair fixation.57 During the verb generation task, subjects generated verbs associated with a visually-displayed noun (‘Generate’) or repeated a visually-displayed noun (‘Repeat’). There were four 30-s blocks per condition and four cross-hair fixation blocks.58

Statistical analysis of clinical and neuropsychological data

Data were analysed using R 3.6.1 and SPSS 27. For demographics, we used Fisher’s exact test, one-way ANOVA and Kruskal–Wallis tests for categorical, continuous parametric and nonparametric variables, respectively. Neuropsychological data were compared via ANCOVA, covarying for age and sex. Comparisons against published norms were attained with one sample t-test. Working memory task performance measures were not normally distributed and were compared via Kruskal–Wallis tests. Across cognitive domains, we corrected for multiple comparisons via the false discovery rate (FDR) procedure.59 Post hoc tests were Bonferroni-corrected.

Functional MRI data: Pre-processing and voxel-based statistics

Functional imaging data were analysed with SPM12 (https://www. fil.ion.ucl.ac.uk/spam/). Images were realigned, normalized to a scanner- and acquisition-specific echo-planar imaging template in Montreal Neurological Institute (MNI) space, resampled to 3 × 3 × 3 mm isotropic voxels and smoothed with a Gaussian kernel of 8 × 8 × 8 mm full-width at half-maximum.60 Individual-level condition-specific effects were derived via general linear models. Task conditions were modelled as 30-s blocks and convolved with the canonical haemodynamic response function. For verbal fluency fMRI, we created activation contrasts associated with generating words. For verb generation fMRI, we subtracted word repetition from word generation. For verbal working memory fMRI, we subtracted verbal monitoring from the 2 Back working memory condition. For visuospatial working memory fMRI, we contrasted the condition with low working memory demand against the active control condition (1–0 Back) and directly compared activation for high and low working memory demand (2–1 Back). Voxel-wise contrast estimates (β weights) were computed with six motion parameters as confound regressors. Scans with a mean framewise displacement > 0.5 mm were discarded from further analysis.61 Further quality checks are detailed in the Supplementary material.

Group analyses were conducted with nonparametric permutation tests using SnPM1362 (http://www.nisox.org/Software/SnPM13/) to attain methodological homogeneity across analytical scales. One-sample permutation t-tests assessed the effects of each task condition per group. Following exploratory permutation-based F-tests, group differences were assessed via two-sample permutation t-tests, all with 10,000 permutations and age and sex as covariates. Comparisons of FLE and TLE included the side of epilepsy as an additional covariate (Fig. 1). Statistical significance was set at two-tailed P<0.05, voxel-wise corrected for family-wise error rate (FWE)63 within pre-specified language, working memory and default-mode (‘task-negative’) regions of interest (ROIs; Fig. 1 and Supplementary material). For completeness, we report group differences for areas outside such ROIs at two-tailed PFWE<0.05, voxel-wise corrected brain-wide.

Functional MRI data: canonical systems and principal gradient

We quantified task effects across seven resting-state functional systems35 (Fig. 2): visual, somatomotor, dorsal attention, salience (or ventral attention), (para)limbic, frontoparietal control and DMN. For each task contrast, we extracted β weights from all parcels of the Schaefer brain atlas64 (200 ROI scale, MNI space) using FSL-6.0.2, averaged β weights across ROIs belonging to a given system and adjusted them for age and sex via multiple regression. We profiled task effects along the principal functional connectivity gradient in surface space (Supplementary material). 

The gradient was computed from resting-state fMRI data of 100 Human Connectome Project (HCP) participants via non-linear dimensionality reduction of surface-registered functional connectivity metrics.66,67 HCP acquisition and preprocessing have been detailed elsewhere.68 The gradient was discretized into 20 equally-sized bins, as previously reported 39,65; cortical locations were assigned to each bin, with sensory/motor regions assigned to the 1st bin and transmodal regions assigned to the 20th bin. For each participant and task contrast, we derived average β weights per bin via a sliding window approach69 and adjusted them for age and sex via multiple regression.

In controls, one-sample permutation t-tests assessed task effects per system or gradient bin. We computed deviation (Z) scores to determine the atypicality of effects in patients [Zpat=(Actpat−μCTR)/σCTR], where μCTR and σCTR correspond to the mean and standard deviation of a systems-level or bin-wise β weight in controls for a given task contrast.70,71 For each system or gradient bin, Z-score deviations from zero in patients were assessed with two-tailed, permutation-based one-sample t-tests. FLE and TLE were compared via permutation-based two-tailed two-sample t-tests. We also assessed global differences between curves of gradient-stratified task effects (area between curves, AbC), using a nonparametric permutation test based on functional data analysis (FDA) techniques72 (Supplementary material). We used 10000 permutations for all tests and reported Cohen’s d effect sizes. P-values were FDR-adjusted for several systems or gradient bins; comparisons reaching uncorrected P<0.05 (Punc) are reported for completeness. Sensitivity analyses probed effects across DMN and frontoparietal control system subdivisions derived from a more fine-grained 17-system parcellation.35

Correlation of fMRI data with cognitive and clinical variables 

Across scales, we assessed correlations of task effects with cognitive performance in all participants21,42 using permutation-based analyses entailing 10,000 permutations. Voxel-based regressions were conducted with SnPM13; age, sex and group were nuisance covariates. Associations between cognitive scores and fMRI metrics were explored within language, working memory and task-negative ROIs (Fig. 1). Effects are reported at two-tailed, voxel-wise PFWE<0.05. For correlations between cognitive scores and task effects across systems or on the gradient, parameterized as age- and sex-adjusted β weights, we employed permutation-based two-tailed product-moment correlations. For working memory task performance measures, which were skewed, we employed permuted rank correlations. Correlations between fMRI activity and clinical variables, such as age at seizure onset, disease duration, seizure frequency, FBTCS history and time since last seizure21,73,74 were separately computed in FLE and TLE to disentangle syndrome-specific effects using SnPM13-based regressions with sex and side of seizure focus as covariates; age was an additional covariate for models including seizure frequency, FBTCS and time since last seizure. Statistical significance was established using the same ROIs as above. For correlations between clinical variables and task effects across systems or gradients, we used two-tailed, permutation-based correlations.

Data availability

Data to reproduce the main group findings are available on NeuroVault (https://identifiers.org/neurovault.collection:13042). Other data are not publicly available due to their containing information that could compromise the privacy of research participants. Example code is available at: https://github.com/lcaciagl/ Language_WM_FLE_vs_TLE.

Results

Neuropsychological data and fMRI task performance

Patients with FLE differed from controls and/or published norms for most cognitive measures (all PFDR<0.001; see Supplementary Table 1 for test scores and associated statistics). Patients with FLE had better performance on naming, verbal learning and verbal recall tests and worse performance on a mental flexibility test than those with TLE (post hoc P<0.05, Bonferroni-corrected). Working memory and verbal fluency were equally impaired in FLE and TLE. Verbal working memory task execution was less accurate in FLE than in controls, but similar between FLE and TLE (>80% median accuracy in both patient groups). For visual working memory, performance in FLE was worse than controls, with more marked differences for higher task difficulty; there were no differences between FLE and TLE. Supplementary Table 1 provides details regarding fMRI task performance scores and associated statistics.

Cognitive fMRI: synopsis

During language tasks, we found reduced frontal activation and reduced deactivation of DMN nodes in FLE compared to controls. During working memory, FLE showed reduced frontoparietal activation, reduced DMN deactivation and global disorganization of task-related recruitment. For visual working memory, we observed a combination of (i) increased frontoparietal activation and less DMN deactivation than controls for low-level task demands, followed by (ii) reduced frontoparietal activation for higher task demands. Patterns of dysfunction in FLE and TLE broadly overlapped; altered DMN deactivation, however, was more evident in FLE, while reduced activation of posterior language areas was more marked in TLE.

The following sections detail these findings. For voxel-based analyses, the figures show regionally unconstrained whole-brain maps and outline corrected as well as uncorrected findings for completeness, by benchmark evidence.75 As detailed above, voxel-based statistical tests focused on effects in prespecified cortical regions, and we only discuss findings surviving voxel-wise FWE correction for multiple comparisons. Statistical details are provided in Supplementary Tables 2–17.

Verbal fluency fMRI

In controls, the verbal fluency task activated fronto-temporoparietal cortices, hippocampus and subcortical regions (Fig. 2); deactivation encompassed DMN areas, including medial prefrontal, medial parietal and angular cortices. Analysis of systems provided an ensemble perspective on these findings, showing activation of frontoparietal control and salience systems (β=0.10/0.08, PFDR= 0.004/0.020), and tendencies for deactivation of the whole DMN (β = −0.06, Punc=0.038). Gradient-based profiling sorted cortical regions according to a sensory-to-transmodal hierarchy, showing: (i) positive effects at the unimodal gradient end, reflecting activation of visual/primary sensory areas; (ii) positive effects along intermediate and right-sided segments, indicating attentional (perceptually-coupled) and high-order executive processing; and (iii) a negative deflection at the transmodal gradient apex, capturing default-mode deactivation (all PFDR<0.05).

At the voxel level (Fig. 3A), patients with FLE had reduced activation of the left middle and inferior frontal gyrus, middle-anterior and middle-posterior temporal areas (PFWE<0.05), and reduced deactivation of bilateral anterior and posterior DMN regions, left posterior temporal and angular gyrus (PFWE<0.05) compared to controls. In TLE, there was reduced left inferior frontal activation and reduced deactivation of bilateral precuneus (PFWE<0.05) compared to controls (Supplementary Fig. 1). Patients with FLE had similar cortical activation to the TLE group, but lesser deactivation of posterior temporal and anterior DMN areas (PFWE<0.05). Across systems (Fig. 3B), there were no corrected differences between FLE and controls; sensitivity analyses across 17 systems highlighted impaired deactivation of DMN and frontoparietal control subdivisions in FLE than controls (DMN-A/DMN-C/control-C: PFDR=0.004/ 0.0025/<0.0001, d=0.51/0.39/0.65; Supplementary Fig. 2). Curves of gradient-based task effects in FLE versus controls (Fig. 3C) showed (i) weaker task activity in intermediate gradient segments; and (ii) an increase at the transmodal apex, which implies lesser deactivation (all PFDR<0.05; d= −0.39 and −0.42 for the intermediate bins, d= 0.54 and 0.64 for the apex bins). Comparisons of TLE and controls and FLE and TLE showed no corrected differences for analyses of systems and gradients.

Verb generation fMRI

In controls, the verb generation task activated fronto-temporoparietal cortices and subcortical areas (Fig. 2D); as distinct from verbal fluency, the left posterior temporal cortex and angular gyrus belonged to the task activation map. Activation involved frontoparietal control, DMN, salience and dorsal attention systems (β=0.12/ 0.06/0.08/0.06, PFDR<0.0001/0.017/0.001/0.017, respectively; Fig. 2E). Gradient profiles (Fig. 2F) indicated extensive activation across the intermediate-to-transmodal segments (all PFDR<0.05).

At the voxel level (Fig. 3D), FLE exhibited reduced left inferior frontal activation and reduced right angular deactivation compared to controls (PFWE<0.05); in TLE, there were widespread frontotemporal-parietal and occipital activation reductions compared to controls (PFWE<0.05; Supplementary Fig. 1). FLE had higher left posterior temporoparietal and bilateral occipital activation and lower deactivation of the right angular gyrus and bilateral precuneus than TLE (all PFWE<0.05). Analysis of systems (Fig. 3E) showed no corrected differences between FLE and controls or TLE; there was lower activity in TLE than controls, mostly encompassing dorsal attention, frontoparietal control and salience systems (all PFDR<0.0001; d= −0.60/−0.70/−0.62). Gradient curves (Fig. 3F) showed one positive deviation at the transmodal apex in FLE compared to controls (Punc=0.017, d=0.32), while TLE differed from controls for global gradient-stratified profiles (FDA, permuted P=0.046) and across most gradient bins (all PFDR<0.05; d range= −0.60 to −0.30). One intermediate bin showed higher task activity in FLE than in TLE at an uncorrected threshold (Punc=0.048, d=0.39).

Verbal working memory fMRI

In controls, the verbal working memory task elicited bilateral frontoparietal activation (Fig. 4A), mapping on dorsal attention and control systems (β=0.20/0.26, PFDR<0.0001; Fig. 4B). Deactivation involved posterior cingulate cortex/precuneus, medial prefrontal and sensorimotor cortices (β= −0.08, PFDR=0.002 for somatomotor system effects). Gradient profiling (Fig. 4C) showed positive shifts along intermediate-to-transmodal segments, implicating attentional and executive processing, and decreases at the DMN apex (all PFDR<0.05).

At the voxel level (Fig. 5A), there was reduced frontoparietal activation and reduced deactivation of DMN areas in FLE versus controls (PFWE<0.05), and only reduced frontoparietal activation in TLE versus controls (PFWE<0.05; Supplementary Fig. 1). FLE showed less deactivation of posterior DMN areas than TLE (PFWE<0.05). Analysis of systems (Fig. 5B) showed lower dorsal attention and frontoparietal control system activity in both FLE (PFDR<0.0001/0.019, d= −0.62/ −0.40) and TLE (PFDR<0.0001/<0.0001, d= −0.82/−0.73) compared to controls. Gradient-stratified profiles (Fig. 5C) showed lower activity along intermediate gradient segments (PFDR<0.05; d range= −0.52 to −0.33) in FLE versus controls, and reduced activity across most gradient bins along with global differences in gradient profiles in TLE versus controls (PFDR<0.05; d range= −0.70 to −0.24; FDA, P= 0.030). There were no differences between FLE and TLE for the analysis of systems and gradients.

Visual working memory fMRI

In controls, the 1–0 Back contrast (Fig. 4D) elicited bilateral frontoparietal activation and deactivation of midline DMN areas. Contrasting high versus low working memory demands (2–1 Back) showed increasing frontoparietal recruitment (Fig. 4G). Analysis of systems (Figs. 4E and H) identified dorsal attention and frontoparietal control system activation (β=0.11/0.09, PFDR<0.0001/0.0002, 1–0 Back; β= 0.08/0.11, PFDR=0.005/<0.0001, 2–1 Back), DMN deactivation for the 1–0 Back contrast (β= −0.05, PFDR=0.023), and somatomotor deactivation for both contrasts (β= −0.08 and −0.06, PFDR<0.0001 and 0.005, 1–0 Back and 2–1 Back). Gradient analyses (Fig. 4F and I) indicated positive activity shifts along its intermediate to transmodal segments and significant decreases at the default-mode apex (all PFDR<0.05).

For voxel-wise 1–0 Back contrast comparisons (Fig. 5D), there was increased parietal and dorsolateral frontal activation as well as reduced deactivation of anterior DMN areas in FLE compared to controls (PFWE<0.05), and reduced deactivation of anterior DMN areas in TLE than controls (PFWE < 0.05; Supplementary Fig. 1). Analysis of systems (Fig. 5E) showed higher frontoparietal control and DMN effects in FLE than controls (PFDR = 0.015/0.026, d = 0.47/0.41) and no significant differences between TLE and controls. Gradient profiles (Fig. 5F) globally differed between FLE and controls (FDA, P = 0.022); bin-wise analyses showed higher task-related effects in FLE across most gradient sections (PFDR< 0.05, d range = 0.31–0.51). In TLE, there was less deactivation than controls at the transmodal apex (PFDR< 0.05, d = 0.48 and 0.55). There were no significant differences between FLE and TLE for voxel-based, system or gradient analyses.

For voxel-wise 2–1 Back contrast analyses (Fig. 5G), both FLE and TLE (Supplementary Fig. 1) showed less frontoparietal activation than controls (PFWE<0.05). Analysis of systems (Fig. 5H) showed pronounced negative systems-level deviations in FLE versus controls, particularly for dorsal attention and frontoparietal control systems (PFDR<0.0001/0.007, d= −0.68/−0.56); similar changes were observed for TLE versus controls (PFDR=0.028/0.055, d= −0.39/−0.33 for dorsal attention and frontoparietal control activity). Gradient-based profiles (Fig. 5I) showed global disorganization of task-related recruitment in FLE compared to controls (FDA, P=0.034), with widespread involvement of intermediate and transmodal gradient segments (all PFDR<0.05; d range = −0.60 to −0.32). There were no suprathreshold differences between TLE and controls for gradient analyses, nor between FLE and TLE for voxel-based, system or gradient analyses.

Correlation of fMRI measures with cognitive performance

For language fMRI tasks, higher inferior frontal activation was associated with higher out-of-scanner verbal fluency and naming scores; naming also positively correlated with lateral temporal activation, particularly during verb generation fMRI (Fig. 6A). Conversely, lesser deactivation of bilateral precuneus during verbal fluency fMRI and right posterior temporal areas during verb generation fMRI related to lesser out-of-scanner fluency and naming performance, respectively (PFWE<0.05). Correlations across systems were limited (rperm =0.16, Punc=0.046, correlation of limbic system activity during verb generation fMRI and naming scores; Fig. 6B). For verbal fluency fMRI, gradient-based effects at the transmodal apex negatively correlated with out-of-scanner letter fluency scores (rperm = −0.17/−0.19, Punc=0.036/0.026).

For verbal working memory fMRI, out-of-scanner digit span scores positively correlated with (i) bilateral frontoparietal activation during verbal working memory (PFWE<0.05; Fig. 6D); (ii) activity across dorsal attention and frontoparietal control systems (sperm =0.31 and 0.33, respectively; both PFDR=0.0007; Fig. 6E); and (iii) task signal across intermediate and transmodal gradient sections (all PFDR< 0.01, sperm range: 0.24–0.31; Fig. 6F). Similar patterns were evidenced for correlations between verbal 2 Back task performance scores (in the scanner) and verbal working memory fMRI activity across (i) dorsal attention/frontoparietal control systems (ρperm =0.23/0.21, PFDR=0.020/0.027); and (ii) intermediate-to-transmodal gradient segments (PFDR<0.05, ρperm range: 0.20–0.23).

For 2-1 Back visual working memory fMRI, visual 2 Back task performance scores (in the scanner) positively correlated with (i) bilateral frontoparietal activation at the voxel level (PFWE<0.05); (ii) activity across dorsal attention and frontoparietal control systems (ρperm =0.37 and 0.39; both PFDR<0.0001); and (iii) task signal across intermediate and transmodal gradient sections (all PFDR< 0.05, ρperm range: 0.20–0.36).

Figure 6 Correlations of functional imaging measures with cognitive performance. Brain renders and sections on the left display statistical maps of nonparametric multiple regressions probing associations between language fMRI (verb generation) and naming scores (A), and between working memory fMRI (verbal task) and digit span scores (D). Cold/warm colour scales refer to negative/positive associations, respectively. Maps are shown at P< 0.005 uncorrected, with an extent threshold of 10 voxels applied for display purposes; colour bars indicate corresponding t-score scales. ^Scatterplots highlight data distribution for the peak voxel within areas highlighted with a black circle; for illustration purposes, we used age- and sex-adjusted (residualized) contrast estimates (β) as measures of task effect. MNI coordinates and P-values are provided in the Supplementary material. The spider plots (B and E) and gradient plots (C and F) show correlation coefficients for associations between cognitive measures (naming/digit span) and task effects (verb generation/verbal working memory) across each system or gradient bin. Example scatterplots highlight data distribution for correlations at the level of one given system or bin; for analyses of systems: ***PFDR<0.01; **PFDR<0.05; *uncorrected P< 0.05; for analyses along the gradient: *PFDR<0.05; ∇uncorrected P<0.05.

For more information:1950477648nn@gmail.com