Indirect structural changes and reduced controllability after temporal lobe epilepsy resection

Abstract

Objective:

To understand the potential behavioral and cognitive effects of mesial temporal resection for temporal lobe epilepsy (TLE) requires a method to characterize network-wide functional alterations caused by a discrete structural disconnection. The objective of this study was to investigate network-wide alterations in brain dynamics of TLE patients before and after surgical resection of the seizure focus using average regional controllability (ARC), a measure of the ability of a node to influence network dynamics.

Methods:

Diffusion weighted imaging (DWI) data was acquired in 27 drug-resistant unilateral mesial TLE patients who underwent selective amygdalohippocampectomy. Imaging data was acquired before and after surgery and a presurgical and postsurgical structural connectome was generated from whole-brain tractography. Edge-wise strength, node strength, and node ARC were compared before and after surgery. Direct and indirect edge-wise strength changes were identified using patient-specific simulated resections. Direct edges were defined as primary edges disconnected by the resection zone itself. Indirect edges were secondary measured edge strength changes. Changes in node strength and ARC were then related to both direct and indirect edge changes.

Results:

We found nodes with significant postsurgical changes in both node strength and ARC surrounding the resection zone (paired t-tests, p<0.05, Bonferroni-corrected). ARC identified additional postsurgical changes in nodes outside of the resection zone within the ipsilateral occipital lobe, which were associated with indirect edge-wise strength changes of the postsurgical network (Fisher’s exact test, p<0.001). These indirect edge-wise changes were facilitated through the “hub” nodes including the thalamus, putamen, insula, and precuneus.

Significance:

Discrete network disconnection from TLE resection results in widespread structural and functional changes not predicted by disconnection alone. These can be well-characterized by dynamic controllability measures such as ARC and may be useful to investigate changes in brain function that may contribute to seizure recurrence and behavioral or cognitive changes after surgery.

Introduction

Temporal lobe epilepsy (TLE) is a neurological disorder characterized by recurrent and unpredictable seizures, often arising from small, focal regions in the hippocampus¹. Surgical resection of the seizure focus is an effective treatment² in approximately two-thirds of drug-resistant patients^3–5. Some potential mechanisms for postsurgical recurrence of seizures include incomplete removal of the seizure focus⁶, additional impairments outside of the seizure focus prior to surgery⁷, or changes in brain function after surgery^8,9. While much effort has been placed on identifying the seizure focus and other epileptogenic regions in the brain prior to surgery to improve surgical outcomes¹⁰, the widespread effects of epilepsy surgery on the brain have yet to be completely quantified. Furthermore, prediction of postsurgical changes prior to the intervention would greatly aid patient management.

To understand and quantify the widespread effect of epilepsy and epilepsy surgery, one must consider that epilepsy is a network disorder that extends beyond the seizure focus^11–13. Thus, we are challenged to characterize network-wide functional alterations caused by a discrete structural disconnection. This requires measures that are capable of characterizing how the structural network is involved in brain function. To address this, recent studies have utilized network control theory^14,15 and communication-based models¹⁶ on structural networks to better relate changes in brain structure and function in TLE. In the framework of network control theory, average regional controllability (ARC)¹⁷ is one inferred measure of brain function derived from structural networks. It represents the ability for each brain region to steer the brain network into different states. ARC uses mathematical models of brain dynamics to provide insights into plausible patterns of functional relationships between regions of a network, which can be used to investigate how these relationships change when the structural network is perturbed¹⁸, such as after surgery.

The objective of this study was to investigate how structural disconnection from surgical resection of the seizure focus results in network-wide alterations in both structure and function in patients with mesial TLE. We utilized structural connectivity, derived from diffusion MRI, to create a whole brain network model consisting of nodes identified by anatomic regions and edge strength between nodes. We then used ARC as a measure of functional brain dynamics at each node. To meet our objective, we addressed four specific questions: 1) How are changes in structural edge strength directly (primary) and indirectly (secondary) related to the disconnection of white matter tracts through the resection cavity? 2) How does node strength, measured as the sum of edge strength to a given node, change after surgery? 3) How does node function, estimated by ARC, change after surgery? and 4) How are pre to postsurgical changes in edge strength (the principal effect of surgery), node strength, and node ARC related? We anticipate that an understanding of structural alterations and brain dynamics secondary to those caused by the direct removal of brain tissue may more completely inform and predict changes in brain function that may contribute to seizure recurrence or other neurocognitive changes after surgery.

Methods

Participants

We enrolled 27 patients (Table 1), who were diagnosed with unilateral mesial TLE using standard clinical assessments including structural MRI, long term video scalp electroencephalography (EEG), seizure semiology, and positron emission tomography (PET) imaging. Patients with suspected bilateral seizure onset or who had identifiable structural abnormalities on MRI outside of the mesial temporal lobe were excluded. All patients underwent a selective amygdalohippocampectomy, a targeted surgical resection of mesial temporal structures that spares temporal neocortex¹⁹, as part of their standard clinical care. In all cases, left TLE patient data were flipped to align with right TLE patients. The use of ipsilateral refers to the hemisphere of seizure onset, while contralateral refers to the hemisphere opposite of seizure onset. In addition, 33 healthy control subjects (Table 1), with no history of neurologic disease, were also enrolled in this study. The controls were not significantly different from patients in age (p=0.24, Wilcoxon rank-sum test) or sex (p=0.41, Chi-squared test). This research was approved by the Vanderbilt University Institutional Review Board. Informed written consent was obtained from all participants.

Table 1.

Participant Demographics

	TLE (N=27)	Healthy Controls (N=33)	P-value
Age (years)	41.0 +/− 12.5	36.9 +/− 13.1	0.24a
Sex (% female)	41.0 (n=11)	51.5	0.41b
Duration of epilepsy (years)	25.5 +/− 17.4
Surgery side (% left)	37.0 (n=10)
Histopathology confirmed mesial temporal sclerosis (%)	96.3 (n=26)
Time from surgery to postop scan (months)	26.3 +/− 16.4
Time from preop scan to surgery (months)	2.3 +/− 3.4
Engel outcome at 1 year postop (%)
Engel I	88.9 (n=24)c
Engel II	7.4 (n=2)
Engel III	3.7 (n=1)
Engel IV	0.0 (n=0)

^aWilcoxon rank-sum test

^bChi-squared test

^cOne patient had only a 6-month outcome score

Imaging

All MRI scans were acquired on a Philips 3T MRI scanner with a 32-channel head coil (Philips Healthcare, Best, Netherlands). For each patient with TLE, one scanning session was performed prior to surgery and one was performed at least one year after surgery. For each healthy control subject, one scanning session was acquired as a baseline and another was acquired as a one-year follow-up. Each scan session consisted of a T1-weighted 3D MRI scan (gradient echo, TR = 9.0 ms, TE = 4.6 ms, 1×1×1 mm³) for tissue segmentation and a diffusion-weighted image (spin echo, multi-slice, TR = 9186 ms, TE = 74 ms, 2.5×2.5×2.5 mm³, 92 gradient directions, b-value=1600 s/mm²) for whole-brain tractography.

Tissue and regional segmentation

A multi-atlas automated brain parcellation²⁰ was used to segment 113 brain regions from the T1 MRI, excluding the left and right hippocampus. Additionally, FreeSurfer 6²¹ was used to segment subfields of the left and right hippocampi to create one anterior and one posterior hippocampal region per hemisphere²², which was then merged with the multi-atlas segmentation. The ipsilateral amygdala and anterior hippocampus were excluded from all analyses, resulting in a total of 115 nodes used as regions of interest. Due to our heterogeneous TLE cohort and the subject-specific nature of resection surgery, we opted for a brain parcellation with broader regions to mitigate the inherent variability within and between subjects. SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) and MATLAB 2020a (MathWorks, Inc, Natick, MA) were used to segment gray matter, white matter, and cerebrospinal fluid to define the gray matter/white matter boundary used in the following tractography pipeline.

Each of the above segmentations were performed in all scans for both healthy control subjects and presurgical TLE patients. For the post-surgical data, the resection mask was defined using ResectVol²³, an automated segmentation pipeline for brain lacunae. First, the resection mask was segmented from the postsurgical T1 MRI. Second, the presurgical T1 MRI was non-linearly registered to the postsurgical T1 MRI to account for shifts in brain tissue surrounding the resection cavity. Third, this registration was applied to the presurgical brain parcellation. Lastly, voxels in the parcellation overlapping with the resection mask were removed to create the postsurgical brain parcellation. This resection mask was also used to create the patient-specific simulated resection data described below. All results were visually checked for accuracy.

Edge and node strength

Diffusion data were first preprocessed using the automated PreQual pipeline²⁴. Briefly, preprocessing included denoising, motion and eddy current correction, B1 field bias correction, and an echo-planar B0 distortion correction²⁵. For subsequent whole-brain tractography, the fiber orientation distribution (FOD) was estimated at each voxel²⁶, the gray matter/white matter interface was used for anatomically-constrained tractography²⁷ in MRTrix 3²⁸. 2×10⁷ streamlines were generated and then reduced to 1×10⁷ and each given a weighting using the spherical deconvolution informed filtering of tractograms (SIFT) and SIFT2 methods²⁹. The structural connectome (SC) was defined as a 115×115 matrix with the edge-wise connection strength computed as the number of streamlines connecting two node pairs, weighted by SIFT2, streamline length, and the inverse of the node sizes. Each SC was then thresholded at 25% density and was then normalized by dividing each element by the largest singular value of connection strength in the matrix to define edge strength. Both pre and postsurgical networks were normalized. All SC graphs were checked to confirm they were fully connected, meaning the thresholding did not disconnect a node or set of nodes from the network. A sensitivity analysis was performed, in which ARC and node strength were calculated for density values from 0 to 100% in 5% increments. The values in all regions stabilized at a density threshold of 25% (Figure S1). Total connection strength of a single node (node strength) was computed as the sum of its edge strength connections to all other brain nodes.

Average regional controllability (ARC)

The analyses in this study model the brain as a linear time-invariant system with nodes connected to one another by edges through the defined SC. Here, the activity level (x) of the nodes, at the next discrete timepoint ($\mathrm{t}+1$), is determined by the SC weighted activity levels of their connected neighbors, plus any additional input energy into the node itself (Equation 1)¹⁷:

$$x\left(t+1\right)=Ax\left(t\right)+B_k{u}_k\left(t\right)$$ (1)

where A is the adjacency matrix defined by the SC, $B_k$ is a list of all nodes in the network, and $u_k$ is any energy input into the network at node k. Here, controllability refers to the ability to drive a dynamic system to a target state with input energy. ARC is defined as the average input energy from the set of control nodes over all possible states¹⁷. This definition is similar to measuring the impulse response of a circuit at every node, and therefore ARC is computed as the trace of the inverse of the controllability Gramian, Trace (${\mathrm{W}}_k^{-1})$, from Equation 2:

$$W_k=\sum _{t=0}^{\mathrm{\infty }}{A}^t{B}_k{B}_k^T{A}^t$$ (2)

The system described in Equation 1 is controllable if and only if ${\mathrm{W}}_{\mathrm{k}}$ is invertible. This ARC was computed for every brain node in each subject. Therefore, a node with high ARC can be interpretated as having more influence over the activity of network dynamics compared to a node with low ARC.

Direct and indirect edge strength changes

We identified edges whose connection goes directly through the resection cavity as edges whose change is directly related to the surgery. These ‘direct’ edges were determined with a simulated resection using each patient’s postsurgical resection volume and presurgical diffusion data. All tractography streamlines within each patient’s presurgical diffusion data were removed if they intersected with the resection volume defined above and the total number of removed streamlines was quantified. Then a new SC matrix was computed for this simulated resection factoring in the change in the total number of streamlines and the changes in region size. We then identified the direct edges that showed strength reduction across the TLE cohort between the baseline presurgical SC and the simulated resection SC, by thresholding for edge-wise strength reductions of at least 8%. This threshold was determined by computing a null model of edge changes within the healthy control cohort between their baseline and one year follow-up scans. The standard deviation of edge changes in this null model was 4.0% and gave a 95% confidence level for all changes within this 8% threshold. Therefore, any change greater than 8% within our simulation resection was considered to be significant.

Next, the measured edge-wise strength reduction across the cohort between the presurgical SC and the post-surgical SC were computed for each of the 6,555 edges in the SC using paired t-tests (p<0.05), with Bonferroni correction for multiple comparisons (α=0.05/6,555). These changes were assumed to contain both direct and indirect edge connections to the resection. Thus, we identified indirect edge-wise connection changes as the measured pre to postsurgical edge changes that were not predicted by the simulated resections. Last, to quantify any time or noise related changes in the SC, the change in edge strength between the baseline and the one year follow up in the healthy controls was computed using the same paired t-tests with Bonferroni correction.

Postsurgical node strength vs. ARC changes

Changes in node strength were compared in the TLE cohort between the presurgical and postsurgical data in 115 nodes using paired t-tests (p<0.05), with Bonferroni correction for multiple comparisons (α=0.05/115). Similarly, changes in ARC were also computed between the presurgical and postsurgical data using paired t-tests with correction. For healthy controls, both ARC and node strength were compared between baseline and one-year follow-up using paired t-tests with correction. In addition, we computed the presurgical to postsurgical node volume changes in any nodes that were found to have strength or ARC differences using paired t-tests with correction.

Relating edge changes to node changes

Lastly, we investigated the relationship between the edge changes and the node changes. To do this we considered all the edge changes (direct and indirect) in the presurgical to postsurgical data. All the nodes associated with these edges were grouped into four types: 1) nodes with both strength and ARC changes (strength + ARC node), 2) nodes with only ARC changes (ARC only node), 3) nodes with only strength changes (strength only node) and 4) ‘other’ nodes that were not identified by significant presurgical to postsurgical node changes of either type. From this network, we computed the number of direct and indirect edges involving each of these four types of nodes. A two-sided Fisher’s Exact test was used to determine whether type of edge (direct or indirect) was associated with type of node (strength + ARC, ARC only, strength only or other) using SPSS 28 (IBM Corp., Armonk, NY).

Data availability

Data supporting the findings of this article are available from the corresponding author upon reasonable request.

Results

Direct and indirect edge strength changes

As expected, the simulated resection showed the most edge strength reductions within the ipsilateral temporal lobe surrounding the resection zone (Figure 1A and B). Each edge change was shared by at least 16 out of 27 patients. The average percentage of streamlines removed by the simulated resections was 1.09% +/− 0.68%. Temporal-temporal connections were the most disrupted, followed by temporal-subcortical, temporal-occipital, and temporal-parietal connections. Each node within the parietal, occipital, and subcortical regions showed only one or two edge changes to nodes within the temporal lobe.

Figure 1. Simulated resection on presurgical data from patient-specific resection volumes.

(A) Resection volume (red) used to simulate a resection on presurgical SC data, shown for a single patient. (B) The nodes and edges that are disconnected by the simulated resection across all patients, defined by an 8% reduction in edge strength. Each edge change was shared across at least 16 out of 27 patients. Nodes are colored by lobe with parietal (blue), occipital (teal), temporal (yellow), and subcortical (maroon).

Measured edge-wise strength changes between the presurgical and postsurgical data were also identified using paired t-tests (Figure 2). Significant edge-wise changes were identified within temporal-temporal, temporal-subcortical, temporal-occipital, and several temporal-parietal connections. Additionally, many occipital-subcortical connections were identified. We then compared the edge-wise strength changes predicted by the simulated resections to the measured postsurgical changes. The edge-wise changes that agree between the simulated resection and postsurgical data are shown in red and are considered the direct edge changes. These connections were located across temporal-temporal, temporal-occipital, temporal-subcortical, and temporal-parietal regions, which are the areas we expect to be directly impacted by the resection. The rest of the measured postsurgical edge-wise changes that were not predicted by simulated resection are shown in black and are considered the indirect edge strength changes. Most of the edge-wise changes that were not predicted were occipital-subcortical connections. Overall, the simulated resections captured 12 of 37 of the total edge-wise changes that we observe in the postsurgical data. No significant edge-wise changes were identified within the healthy control cohort between the baseline and one-year follow-up scans.

Figure 2. Measured postsurgical edge-wise changes and identification of direct vs. indirect edges in the TLE cohort.

All significant postsurgical edge-wise changes (paired t-tests, p<0.05, Bonferroni-corrected), with direct (red) and indirect (black) edge connections. Direct edges were defined as the edges that were changed in the simulated resection (Figure 1). Together, the direct and indirect edge changes account for all measured pre to postsurgical edge-wise changes. Therefore, the indirect edges were the measured pre to postsurgical edge-wise changes that were not predicted by the simulated resections.

Postsurgical node strength and ARC changes

Changes in node strength and ARC were first compared in the TLE cohort between the presurgical and postsurgical data using paired t-tests. The ipsilateral anterior hippocampus and amygdala were excluded from the analyses as they are assumed to be completely removed by the resection. Significant reductions in node strength and ARC (Figure 3) overlapped in the ipsilateral temporal lobe within the parahippocampal gyrus, inferior temporal gyrus, middle temporal gyrus, and temporal pole. The ipsilateral lingual gyrus was the only other node to show both ARC and node strength reduction. ARC captured more widespread postsurgical changes mostly in the occipital lobe within the calcarine cortex, occipital fusiform gyrus, inferior occipital gyrus, and cuneus. Additionally, significant postsurgical ARC reduction was found in the fusiform gyrus. Node strength demonstrated a postsurgical change in one node not identified by ARC, the planum temporale. This was a node strength increase, which was unique from all the other changes. The significant node strength and ARC results are summarized in Table 2. No significant strength or ARC changes were identified in the contralateral hemisphere. No significant changes were identified in any nodes within the healthy control cohort between the baseline and one-year follow-up scans. The parahippocampal gyrus and middle temporal gyrus showed significant change in volume size, among the nodes with strength and/or ARC change. All other nodes experienced small or zero change in volume.

Figure 3. Measured presurgical to postsurgical node strength and ARC changes in the TLE cohort.

Significant presurgical to postsurgical node strength and ARC changes were identified using paired t-test (p<0.05, Bonferroni-corrected). Nodes with only strength change are shown in blue, nodes with only ARC change are shown in light blue, and nodes with both strength and ARC change are shown in maroon. No significant changes were identified in the contralateral hemisphere.

Table 2.

Nodes with significant presurgical to postsurgical changes in strength and ARC within the TLE cohort

Lobe	Region Name	Strength Change (% ± STD)	ARC Change (% ± STD)	Volume Change (% ± STD)
Temporal	Parahippocampal Gyrusa	−11.1 ± 10.5	−14.8 ± 12.1	−24.6 ± 19.6*
Temporal	Inferior Temporal Gyrusa	−27.6 ± 14.4	−11.1 ± 12.0	−2.5 ± 3.1
Temporal	Middle Temporal Gyrusa	−15.2 ± 13.5	−10.3 ± 9.7	−6.1 ± 5.1*
Temporal	Temporal Polea	−79.0 ± 22.6	−9.1 ± 7.9	−0.3 ± 1.2
Temporal	Planum Temporale	11.4 ± 11.3	-	−0.01 ± 0.1
Temporal	Fusiform Gyrus	-	−6.9 ± 6.9	−3.0 ± 4.2
Occipital	Lingual Gyrusa	−21.4 ± 24.6	−3.3 ± 3.3	−0.2 ± 0.5
Occipital	Calcarine Cortex	-	−15.0 ± 7.4	0.0
Occipital	Occipital Fusiform Gyrus	-	−17.2 ± 14.4	−0.2 ± 1.1
Occipital	Inferior Occipital Gyrus	-	−1.8 ± 1.3	−0.03 ± 0.1
Occipital	Cuneus	-	−1.7 ± 2.1	0.0

paired t-tests, p<0.05, Bonferroni corrected (α = 0.05/115)

^aNodes with both strength and ARC change

- No significant difference

^*Significant volume change

% = mean percent change

STD = standard deviation

Indirect structural changes influence postsurgical ARC

All direct edges (12/12) connect to at least one strength + ARC node, meaning the node had both significant strength and ARC changes after surgery compared to presurgery. Of these, 3/12 also connect to another strength + ARC node, 1/12 connect to an ARC only node and 7 connect to other nodes without strength or ARC changes. Of the 25 indirect edges, 10/25 connect to at least one strength + ARC node, 11/25 connect to at least one ARC only node, with the remaining 4 connecting to other nodes. The Fisher’s exact test showed a significant association between edge type and node type (p<0.001), where direct edges are more associated with at least one strength + ARC node and indirect edges are more associated with ARC only nodes.

Figure 4 depicts all the direct and indirect edges and their associated nodes. In this figure, 9 of the ‘other’ nodes form edges to the strength + ARC nodes and to the ARC only nodes. We consider these ‘other’ nodes as ‘hub’ nodes, which link the other two types of nodes, and include the posterior hippocampus, anterior insula, posterior insula, thalamus, putamen, superior temporal gyrus, planum polare, middle occipital gyrus, and posterior cingulate gyrus all ipsilateral to the resection. Therefore, we see that the strength + ARC nodes are connected to the hub nodes mostly via direct edges, while the hub nodes are connected to the ARC only nodes via mostly indirect edges.

Figure 4. Indirect edge influence on postsurgical ARC.

Graphical representation of the postsurgical edge-wise changes separated by direct (red) and indirect (black) edges, and nodes grouped by significant changes in both strength and ARC (maroon), ARC only (light blue), hub nodes (green), and remaining nodes (orange). Hub nodes form connection between strength + ARC nodes and ARC only nodes. ARC only nodes were found to be associated with indirect edges (Fisher’s exact test, p<0.001). STR=strength, PHG=Parahippocampal Gyrus, ITG=Inferior Temporal Gyrus, MTG=Middle Temporal Gyrus, TMP=Temporal Pole, LiG=Lingual Gyrus, FuG=Fusiform Gyrus, Calc = Calcarine Cortex, OFuG=Occipital Fusiform Gyrus, IOG=Inferior Occipital Gyrus, Cun=Cuneus, STG=Superior Temporal Gyrus, Put=Putamen, PP=Planum Polare, Thal=Thalamus, Pins=Posterior Insula, PHip=Posterior Hippocampus, PCu=Precuneus, MOG=Middle Occipital Gyrus, AIns=Anterior Insual, AnG=Angular Gyrus, SOG=Superior Occipital Gyrus, cSOG=Contralateral Superior Occipital Gyrus.

Discussion

In this study, we identified structural connectivity changes after TLE resection surgery, and used these to estimate network-wide dynamic functional alterations measured by ARC. There were two primary findings. First, widespread structural changes were detected beyond those expected by direct surgical disconnection as determined by simulated resection. The changes identified by simulated resection on presurgical data were defined as the direct edge strength changes, while indirect strength changes were defined as the additional measured significant edge-wise changes in the postsurgical data that were not predicted by simulated resections. Second, functional changes estimated from the structural network using ARC were quantified and their relationship to direct and indirect strength changes were characterized. By doing so we were able to isolate a set of hub nodes that form a link between nodes with strength and ARC changes via direct and indirect edges, and nodes with only ARC changes via indirect edges. Thus, ARC was able to estimate functional changes resulting from both direct and indirect disruptions after surgery.

From the pre to postsurgical edge strength changes, we estimated node strength and functional ARC changes after surgery. Node strength is a measure of a node’s total connection to its immediate neighborhood. ARC measures a node’s influence over network activity beyond just its immediate neighborhood. Here, strength decreases were interpreted as lowered levels of connection between neighboring nodes. ARC decreases were interpreted as lowered levels of influence over the activity of network dynamics. In TLE resection surgery, the seizure onset zone is removed to stop the spread of irregular electrical activity throughout the brain. Therefore, in our results, node strength and ARC agreed in areas of large disruption (e.g., surrounding the resection zone). However, ARC identified more changes outside of the resection zone, primarily throughout the occipital lobe, that could not be detected by node strength changes after surgery. Nodes with only significant ARC changes from pre to postsurgery were more associated with indirect edge-wise strength changes. These indirect edge changes along with identified intermediary connections through hub nodes demonstrates that changes in brain function can occur secondary to the surgical resection through a pattern of small edge-wise strength changes not directly impacted by surgery.

Specifically, our results show widespread ARC reductions throughout the occipital lobe, meaning these nodes have less influence over network activity. ARC was used an inferred measure of function based on structural diffusion data, and recent work has established a correlation between reductions in control energy and hypometabolism in a TLE cohort³⁴. The occipital lobe has been shown to be related to both structural and functional changes after TLE resection surgery. TLE patients exhibit cortical thinning throughout the temporal and occipital lobes^35,36 and a follow-up study has shown that reduced cortical thinning within the occipital lobe is related to seizure-free outcomes³⁷. Studies have also shown disrupted white matter integrity of temporal-occipital connections in TLE^31,38 and postsurgical disruptions to the inferior longitudinal fasciculus, a white matter bundle between the temporal and occipital lobes involved in visual processing³⁹. Other studies have identified functional connectivity changes using functional MRI⁹ and hypometabolism in the occipital lobe using positron emission topography³³. The reductions in ARC throughout the occipital lobe may also be related to visual field defects, a possible side-effect of TLE resection surgery^40,41.

The biological underpinnings and functional consequences of structural white matter reorganization after surgery is crucial for understanding seizure freedom and other neuropsychiatric outcomes after surgery. However, there still exists a wide gap in our ability to relate structure and function. Several studies have demonstrated the impact of postsurgical white matter changes on verbal fluency³⁰, visual field defects³¹, and seizure freedom³². The use of fractional or quantitative anisotropy and static graph measures in these studies are unable to detail any direct links between structure and function. Here, the presented ARC methodology seeks to narrow this gap by demonstrating how small structural edge changes within a network can influence functional nodal changes, even outside of the resection zone. This goal is accomplished by utilizing a dynamic model of brain activity on the structural network that provides a whole-brain picture of how each node operates within the network while maintaining a sensitivity to how individual edge changes influence node-to-node interactions. We anticipate that the use of structural measures that approximate functional activity, such as ARC, will better integrate with actual functional measures used in TLE patient management and together will provide a better understanding of treatment outcomes.

This study leveraged the use of simulated resections alongside pre and postsurgical measures to account for several confounding factors and to offer a delineation between expected versus unexpected changes resulting from resection surgery. One confounding factor is the use of tractography and SC computations between pre and postsurgery. The simulations showed a total impact of approximately one percent of all streamlines are directly impacted by resection, which would result in small differences in how postsurgical connectivity is defined. We anticipate this confound would become more influential for larger resections. Similarly, changes in region size must also be accounted for. Additionally, there is a question of whether any reduction in edge strength inherently results in reduced ARC. The important component in answering that question lies in the fact that edge changes occur between pairs of regions. One side of the region pair may experience larger ARC changes than the other side. Our goal with this study was to show that ARC can characterize the complex array of small edge changes that appear as random alterations atop the expected primary edge changes resulting from surgery.

Limitations

One important consideration for this study is the narrow patient cohort, both a limited sample size and a relatively homogeneous cohort of mesial TLE patients who received selective amygdalohippocampectomies. All patients, except for one, had mesial temporal sclerosis. We also combined left-sided and right-sided TLE patients into a single cohort. We acknowledge that differences exist between left and right sided TLE patients, however, the pre to postsurgical patterns of changes we investigated, both node strength and ARC, were consistent across these two populations. Second, the analyses include a postsurgical scan at an average of two years after surgery, and therefore cannot disambiguate between structural changes that occur immediately due to resection and those changes that evolve or continue to evolve over time. Third, clinical outcomes were not specifically investigated in this study. While most patients had Engel I⁴² outcome at one-year postsurgery, outcomes can vary across the time of these post-surgical assessments. Fourth, the ARC relies on a simplified model of brain activity, but this provides a closed-form solution that is easier to compute and interpret compared to advanced, nonlinear models with simulated brain activity. Additionally, all graph metrics computed on structural connectomes are constrained by the inherent limitations and anatomical accuracy of tractography derived from diffusion imaging.

The goal of our simulated resections was to highlight the predicted changes that would occur directly from surgery, but these models relied on resection cavities defined one to five years after surgery. This included the approach corridor to reach the mesial temporal structures, in which tissue is removed but also can be stretched or displaced. Therefore, the resection cavity we identified may not be the whole resection or it may overestimate resection in other areas due to tissue movement. However, small variations in possible resection definitions still cannot account for all of the indirect edge-wise changes, 25 of 37 total edge changes, observed in the measured postsurgical data. Therefore, these indirect changes after surgery cannot be easily predicted from the resection alone. We are confident that these additional changes are occurring uniquely in our TLE cohort after surgery because we do not observe any significant edge changes in healthy controls over a similar time period.

Conclusions

In this study, we demonstrated that dynamic measures of brain activity, such as ARC, derived from structural connectomes can capture indirect changes in function after TLE resection surgery that are not identified by static graph measures, such as node strength. A majority of the edge-wise strength changes within the postsurgical networks contributing to the ARC changes were beyond those that could be predicted by the impact of the resection volume alone. Thus, ARC may provide a better network-level picture of the functional consequences of small underlying structural network connection changes either directly or indirectly related to resection. There may be future clinical utility using ARC to predict the consequence of these indirect connection disruptions in individual patients and relate them to expected seizure outcomes and neurocognitive function. Overall, this study provides a methodology to relate functional changes to unanticipated widespread structural changes after TLE resection surgery.

Key Points.

• Temporal resection results in widespread ipsilateral strength and controllability changes throughout the ipsilateral hemisphere

• Simulated resections show a majority of the postsurgical edge-wise strength changes, 25 of 37, were not directly caused by the resection

• Indirect ARC changes after resection are facilitated by “hub”’ nodes including thalamus, insula and putamen