Resting-state hippocampal networks related to language processing reveal unique patterns in temporal lobe epilepsy

Abstract

Objective:

Patients with temporal lobe epilepsy (TLE) commonly experience a broad range of language impairments. These deficits are thought to arise from repeated seizure activity that damages language regions. However, connectivity between the seizure onset region in the hippocampus and regions related to language processing has rarely been studied, and could also have a strong impact on language function. The purpose of this study was to use resting-state functional connectivity (FC) measures to assess hippocampal network patterns and their relation to language abilities in patients with right TLE (RLTE), left TLE (LTLE), and healthy controls.

Methods:

Presurgical resting state 3T functional MRI data were acquired from 40 mesial TLE patients (27 RTLE, 13 LTLE) and 54 controls. The regions of interest were the anterior and posterior bilateral hippocampi and eleven regions grouped by frontal or temporo-parietal locations, including large areas of language-related cortex. FC values were computed with the right/left anterior and posterior hippocampi as the seeds and frontal and temporo-parietal regions as targets. Resting-state lateralization indices were also calculated (LI-Rest), and all FC measures were correlated to neuropsychological language scores and measures related to manifestation of epilepsy including age of onset, duration of disease, monthly seizure frequency, and hippocampal volume.

Results:

We found significant group differences between the anterior hippocampi and temporo-parietal regions closest to the seizure focus, in which RTLE and LTLE showed stronger connectivity to their contralateral hippocampus, while controls showed similar connectivity to both hippocampi. In addition, LI-Rest demonstrated significantly more right lateralization in LTLE compared to RTLE for temporo-parietal regions only. In LTLE, we found significant associations between stronger hippocampal network resting-state FC and later age of onset and decreased left anterior hippocampal volume.

Significance:

The results of our study indicate that the presence of TLE impacts hippocampal-temporo-parietal networks relevant to language processing.

1. Introduction

A wide variety of language deficits are present in patients with temporal lobe epilepsy (TLE), where seizures originate in the hippocampus and surrounding structures [1–4]. The most common language deficits involve naming and generative fluency [5], though impairments in spontaneous speech, discourse production, and reading have also been observed [6]. It’s estimated that up to a third of all TLE patients have impairments in more than one language function [6], while left TLE (LTLE) patients often have more difficulty than right TLE (RTLE). For example, up to 75% of left TLE patients show deficits on auditory naming compared to just 19% of RTLE [6]. These deficits are generally thought to arise from repeated seizure activity that damages temporal and frontal regions crucial to language processing [5]. Specifically, for the most common language deficits, this would include damage to the dominant posterior inferior temporal gyrus, the left fusiform gyrus, and more anterior temporal regions, depending on the naming category [5, 7], and damage to a broader network of temporal, frontal, and parietal regions depending on the type of fluency measured [8].

In accordance with this view of propagated damage to language regions, studies have found that reorganization of the neural language network away from the seizure onset zone is associated with better language abilities post-surgically, and it tends to occur more frequently with earlier age of onset of epilepsy [5]. This reorganization can take the form of either shifting language dominance from the left hemisphere to the right hemisphere, or shifting language processing commonly associated with temporal regions to more frontal regions [9]. These studies suggest that determining the general architecture and reorganization of language networks in the frontal and temporal regions in TLE patients is crucial to understand how the presence and progression of temporal lobe epilepsy leads to language deficits. Furthermore, the hippocampal seizure focus itself may also contribute to language deficits, as evidenced by greater naming and fluency deficits in patients with hippocampal sclerosis [10, 11]. Although the hippocampus has been thought to be primarily involved in memory, more recent research shows evidence of the direct role of the hippocampus during language processing [12–16], and the results of Pu et al. (2020) [16] suggest that theta phase coupling between the hippocampus and a temporal language region may link the memory and language systems to facilitate language comprehension. Despite this new evidence, the impact of the hippocampi on networks relevant for language processing in TLE patients and how it relates to language deficits remains unknown.

Measuring functional connectivity (FC) from functional Magnetic Resonance Imaging (fMRI) offers the unique ability to noninvasively assess the organization of networks in the brain through their temporal correlations. At present, only a handful of all FC studies in TLE have measured FC of the networks related to language processing from resting-state or task designs [17–24]. In the only studies thus far to measure resting-state FC between the hippocampus and language-related regions in TLE, Roger et al. (2019) [17] found decreased connectivity between the hippocampus and fronto-temporo-parietal regions, while Doucet et al. (2015) [18] found that a graph theory measure of FC from the contralateral hippocampus to regions across the whole brain significantly predicted post-surgical language outcomes in both RTLE and LTLE patients – indicating that FC between the hippocampus and language network may be an important factor affecting language abilities in TLE. However, due to the inclusion of the broad cortex involved in memory processing in Roger et al. (2019) [17], and the whole-brain graph theory measurement in the Doucet et al. (2015) [18] study, knowledge of the connectivity specifically between the hippocampus and language-related networks remains limited. Expanding on these findings by measuring FC from the hippocampi to broad areas of frontal, temporal, and parietal cortex is essential to fully understand the organization of language-related networks in TLE. This may be especially important in TLE patients, given that reorganization often means they do not exhibit the same language region locations and functions as healthy controls [25,26]. Further work is also needed to determine whether the anterior vs. posterior portion of the hippocampus plays a more significant role in language deficits, which could be crucial to surgical planning, as the anterior section is most commonly resected in TLE surgeries while the posterior extent of the resection is more variable [27].

The use of FC measures to quantify hemispheric dominance may also offer additional information to further understand the architecture of language networks in TLE patients. Previous studies had varying success in measuring language lateralization from resting-state FC [28–30], but, to date, no study has measured a laterality index (LI) using FC involving a hippocampal seed, which may be a more informative laterality measure in TLE patients since it could indicate the role of the seizure focus in the larger language network.

With growing evidence supporting the direct role of the hippocampus in language, studies measuring resting-state FC between hippocampi and language-related cortex are needed to better understand hippocampal networks related to language processing and how they may affect language abilities in TLE. The purpose of the present study was to: 1) assess and compare resting-state FC patterns between the bilateral anterior and posterior hippocampi and broad areas of frontal and temporo-parietal cortex, including language cortex, in mesial TLE patients compared to healthy controls, as well as FC patterns of hemispheric dominance at rest, 2) determine whether these FC patterns are different in patients whose seizures originate in the right vs. left temporal lobe, and 3) examine relationships between hippocampal-frontal and hippocampal-temporo-parietal FC and measures of epilepsy and language ability. The results of this study add crucial insight into the neurobiology of language-related networks and their interaction with seizure foci in TLE patients.

2. Methods

2.1. Subjects

Patients with unilateral mesial TLE were recruited from the Vanderbilt University Medical Center Epilepsy Program during presurgical evaluation. The diagnosis was confirmed using seizure semiology profiles, structural MRI, electroencephalography (EEG) monitoring, and positron emission tomography (PET). Patients were excluded if they showed evidence of structural abnormalities, with the exception of hippocampal sclerosis. The final sample included 27 patients with RTLE, 13 patients with LTLE, and 54 healthy controls. The study was approved by the Vanderbilt Institutional Review Board (IRB) in accordance with the Declaration of Helsinki for experiments involving humans and informed written consent was obtained from each subject.

2.2. Image Acquisition and Processing

Resting-state functional MRI and structural MRI imaging were acquired using a 3T MRI scanner with a 32-channel head coil (Philips Healthcare, Best, The Netherlands). The fMRI images were T2*-weighted with a field of view (FOV) of 240 mm, 80x80 matrix, echo time (TE) of 35 ms, and repetition time (TR) of 2 s. Thirty-four slices were acquired for each of 300 total volumes with a slice thickness of 3.5 mm and 0.5 mm gap. Subjects were instructed to keep their eyes closed for the duration of the ten minute resting-state scan. Structural imaging included one high-resolution three-dimensional (3D) T1-weighted whole brain volume (1 x 1 x 1 mm³), and one high-resolution two-dimensional (2D) T1-weighted axial whole brain volume (1 x 1 x 4 mm³). Across all scans, physiological monitoring of cardiac and respiratory cycles were recorded (500 Hz sampling rate) using the pulse oximeter and respiratory belt. Processing of the fMRI images was completed using SPM8 software (https://www.fil.ion.ucl.ac.uk/spm/software/spm8/) and custom scripts in MATLAB (The MathWorks, Inc, Natick, MA, U.S.A.). The processing steps were as follows: slice timing correction, motion correction, physiological noise correction following the RETROICOR method [31], spatial normalization to the Montreal Neurological Institute template, and spatial smoothing using a 6 x 6 x 6 mm³ full-width half maximum (FWHM) gaussian kernel.

2.3. Resting-State Functional Connectivity (FC) Analysis

The regions of interest chosen were the anterior and posterior bilateral hippocampi and eleven regions located in the frontal, temporal, and parietal lobes, including many language regions (see Figure 1). The regions were segmented using a multiatlas segmentation algorithm [32,33] which creates gyral-based ROIs based on the parcellation protocol from Neuromorphometrics (http://neuromorphometrics.com/ParcellationProtocol_2010-04-05.PDF). However, because this atlas segments the entire hippocampi and not subregions, we used the FreeSurfer v. 6.0.0 [34] to segment hippocampal sub-fields and then consolidated these into anterior and posterior regions according to McHugo et al. (2018) [35] and Plassard et al. (2017) [36]. As mentioned previously in the Introduction, we sought to expand on previous findings by measuring broad areas of frontal and temporo-parietal cortex, given that the language network is vastly complex and involves many frontal, temporal, and parietal regions all involved in different stages of language processing [37]. Including broad areas of frontal and temporo-parietal cortex also allowed for differences that often arise in the reorganization of language regions in TLE patients compared to controls [25, 26], as well as the ability to find potential compensatory FC patterns that could be relevant in why TLE patients less commonly exhibit deficits in the language domains of comprehension or articulation and prosody. Therefore, the present work aimed to assess a holistic view of the hippocampal networks relevant for language processing in TLE patients to provide more information than could be gained from solely focusing on a few frontal and temporal language regions, often corresponding to Broca and Wernicke’s areas.

Figure 1.

Regions of interest included in functional connectivity (FC) analyses. The bilateral hippocampi were segmented into anterior (green) and posterior (red) sections. Eleven regions were segmented from frontal, temporal, and parietal lobes. FC was measured with anterior or posterior hippocampus as the seed to each region in ipsilateral and contralateral hemispheres, and then averaged into single FC values to all frontal regions or temporo-parietal regions together, given significantly correlated FC values within frontal and temporo-parietal regions (see Results 3.1). This resulted in four FC values from the anterior and posterior sections of both the right and left hippocampus (sixteen values total). The bottom three diagrams illustrate eight FC values as arrows leading from the hippocampi (four from each of the anterior/posterior hippocampi), with the yellow lightning bolt indicating seizure onset (ipsilateral hippocampus) for RTLE and LTLE patients that may interfere with FC. The frontal regions (purple) included the middle frontal gyrus, orbital inferior frontal gyrus, triangular inferior frontal gyrus. The temporo-parietal regions (blue) included the inferior temporal gyrus, middle temporal gyrus, planum polare, planum temporale, superior temporal gyrus, transverse temporal gyrus, angular gyrus, supramarginal gyrus. Note that our ROIs were gyral-based regions, in which all cortex was encompassed on the particular gyrus, extending to the deepest point of the adjacent sulcus on both sides.

The segmentation created gyral-based regions, in which the regions of interest encompassed all cortex on the particular gyrus, extending to the deepest point of the adjacent sulcus on both sides. For example, our superior temporal gyrus region includes the superior bank of the superior temporal sulcus, and the middle temporal gyrus region includes the inferior bank of the superior temporal sulcus. In total, the frontal regions included the middle frontal gyrus, orbital inferior frontal gyrus, and triangular inferior frontal gyrus, while the temporo-parietal regions included the inferior temporal gyrus, middle temporal gyrus, planum polare, planum temporale, superior temporal gyrus, transverse temporal gyrus, angular gyrus, and supramarginal gyrus.

The fMRI resting-state time series were band-pass filtered from 0.0067-0.1 Hz. Functional connectivity between two regions was measured as the partial Pearson correlation between the time series, with the six motion time series as covariates. All correlation coefficients were converted to Fisher Z values using the Fisher Z transform [38]. FC values were computed between the hippocampus sections to each region, and then averaged into single FC values for the anterior and posterior hippocampus to all frontal regions, and averaged into single FC values for all temporo-parietal regions, given significantly correlated FC values within the frontal and temporo-parietal regions (see Results 3.1). This resulted in a total of sixteen FC values, in which the seed region was always the anterior or posterior section of the right or left hippocampus. Each hippocampus section had four FC values, including connectivity to its ipsilateral frontal and temporo-parietal regions, and to its contralateral frontal and temporo-parietal regions (see Figure 1).

2.4. Resting-State Laterality Index (LI-Rest)

Resting-state lateralization indices were calculated using the Quantitative Intrinsic Laterality Index (Liu et al., 2009 [39]; see modified equation below) using the regional resting-state FC values explained above, where the seed was always the anterior or posterior hippocampus and the target was the frontal or temporo-parietal regions. This resulted in four LI-rest measures total: 1) anterior hippocampus to frontal regions, 2) posterior hippocampus to frontal regions, 3) anterior hippocampus to temporo-parietal regions, 4) posterior hippocampus to temporo-parietal regions.

$$(({\text{L}}_{\text{h}}{\text{L}}_{\mathrm{lang}}-{\text{R}}_{\text{h}}{\text{L}}_{\mathrm{lang}})-({\text{R}}_{\text{h}}{\text{R}}_{\mathrm{lang}}-{\text{L}}_{\text{h}}{\text{R}}_{\mathrm{lang}}))/(|{\text{L}}_{\text{h}}{\text{L}}_{\mathrm{lang}}|+|{\text{L}}_{\text{h}}{\text{R}}_{\mathrm{lang}}|+|{\text{R}}_{\text{h}}{\text{R}}_{\mathrm{lang}}|+|{\text{R}}_{\text{h}}{\text{L}}_{\mathrm{lang}}|)$$

In the above equation, L and R denote the left and right hemisphere, respectively, with L_h and R_h indicating the left and right hippocampus, and L_lang and R_lang indicating the left and right regions. The equation results in an LI-Rest value between −1 and 1, where values closer to −1 are more right lateralized, and values closer to 1 are more left lateralized.

2.5. Epilepsy Measures

We assessed four disease parameters related to the nature and severity of TLE obtained with consent from the patients’ medical record. These included the age of onset in years, the duration of disease in years at the time of their MRI, and total monthly seizure frequency of all types (including focal and generalized seizures originating in the temporal lobe, and both aware or impaired awareness seizures). We also measured hippocampal volume, given that atrophy in the hippocampi is often representative of the nature and severity of disease [40,41]. We estimated the gray matter volume of the right and left anterior and posterior hippocampi using FreeSurfer v. 6.0.0 [34].

2.6. Language Tests

Neuropsychological language tests were obtained as part of presurgical evaluations by a licensed neuropsychologist (MLJ). The language tests included a test of naming ability, the Neuropsychological Assessment Battery (NAB) [42], and two word generation tasks: categories (Word Generation-Categories) and letters (Word Generation-Letters), taken from the Controlled Oral Word Association Test (COWAT) [43]. The T-scores from all language tests were used in order to normalize scores based on age and education. In addition, a broad measure of overall language function was included using standard scores (z-scores) from the Verbal Comprehension Index (VCI) from the Wechsler Adult Intelligence Scale-IV (WAIS-IV) [44] or the Wechsler Abbreviated Scale of Intelligence (WASI) [45].

2.7. Statistical Analyses

We compared differences in resting-state hippocampal language-related networks FC patterns in controls, RTLE, and LTLE groups using mixed repeated measures ANOVAs. Our analyses tested whether the anterior vs. posterior hippocampi showed stronger FC to frontal and temporo-parietal regions and whether this differed by group (controls vs. RTLE vs. LTLE). We also compared whether TLE patients exhibited FC pattern differences when the seed was the ipsilateral hippocampus, and whether this relationship changed when the target was frontal vs. temporo-parietal regions. LI-Rest values were also compared across groups using mixed repeated measures ANOVA. For this analysis, we sought to determine if LI-Rest was more or less lateralized when the seed was the anterior vs. posterior hippocampus, and if the groups showed different lateralization patterns. In addition, we compared whether groups differed on demographic variables (age), epilepsy measures (TLE only), or neuropsychological language tests (TLE only) to further understand our study sample.

In order to assess the relationship of the networks to current language abilities, we analyzed Pearson r correlations between neuropsychological language tests and resting-state FC and LI-Rest. We also analyzed Spearman rank-order correlations (due to skewed nature of the measures) between FC and LI-Rest and the epilepsy measures age of onset of epilepsy, duration of disease, monthly seizure frequency, and hippocampal volume to identify potential disease parameters that may contribute to the reorganization of hippocampal networks.

Multiple comparisons were controlled across all t-tests and correlations using the Benjamini Hochberg False Discovery Rate with an FDR rate (q) of 5% for each family of tests. Families consisted of 16 tests (resting-state FC) or 4 tests (LI-Rest) at a time for each neuropsychological score (four scores) or epilepsy measure (seven measures). For repeated-measures ANOVA, multiple corrections were not implemented as the number of post-hoc tests was kept small, only following-up on two significant interaction effects.

3. Results

Table 1 presents demographic features of the sample along with the epilepsy measures and neuropsychological language scores for RTLE and LTLE.

Table 1.

Sample Demographics and Neuropsychological Language Scores

	Controls	RTLE	LTLE
Demographics				Sig.
Age (years)	37 (13.7)	39.4 (10.6)	37.6 (15.2)	N.S.
Handedness	6 LH	5 LH	2 LH	N/A
% Female	42.60%	55.60%	30.80%	N/A
Epilepsy Measures				Sig.
Age of Onset (years)	N/A	21.01 (13.56)	14.13 (15.43)	N.S.
Duration of Disease (years)	N/A	19.42 (12.97)	23.38 (16.47)	N.S.
Monthly Seizure Frequency	N/A	27.47 (57.88)	14.49 (19.22)	N.S.
Hippocampal volume mm3	N/A	Left Anterior: 1577.46 (230.15)	Left Anterior: 1222.98 (219.93)	p < 0.001
		Left Posterior: 1544.02 (204.82)	Left Posterior: 1200.71 (201.88)	p < 0.001
		Right Anterior: 1405.15 (391.49)	Right Anterior: 1614.87 (267.37)	N.S.
		Right Posterior: 1360.33 (366.44)	Right Posterior: 1495.74 (232.31)	N.S.
Presence of Mesial Temporal Sclerosis (MTS)	N/A	11 of 27 patients	12 of 13 patients	N/A
Neuropsychological Language Tests				Sig.
NAB Naming Test (T-Score)	N/A	51.24 (8.50; n = 20)	40.78 (15.50; n = 9)	N.S.
COWAT Word Generation-Categories (T-Score)	N/A	42.04 (10.12; n =23)	37.09 (8.47; n = 12)	N.S.
COWAT Word Generation-Letters (T-Score)	N/A	43.13 (2.81; n =23)	40.15 (17.47; n =13)	N.S.
WAIS/WASI Verbal Comprehension Index (Z-Score)	N/A	102.08 (13.40; n = 24)	85.50 (17.51; n = 10)	p = 0.02

All data presented in the format Mean (SD).

LH = Left-handed; N.S. = Not Significant; N/A = Not Applicable (data not applicable to control group or sig. test not run)

3.1. Resting-State Regional Functional Connectivity

Across all controls and TLE patients, there were highly significant co-linear relationships within the frontal regions and within the temporo-parietal regions (p < .001). The average Pearson r values between the three frontal regions from each of the four hippocampal seeds was r = 0.66 (SD = 0.11), while the average Pearson r value between the eight temporo-parietal regions from each of the hippocampal seeds was r = 0.49 (SD = 0.19). As such, rather than reporting redundant FC results from the hippocampal seeds to each of the individual eleven regions, we present the averaged FC values across all frontal regions together, and across all temporo-parietal regions together.

3.1.1. FC of Anterior vs. Posterior Hippocampi

Across both hemispheres and for all three groups, FC from the bilateral anterior hippocampi was significantly stronger than the bilateral posterior hippocampus to both the frontal regions (F(1, 91) = 41.78, p < 0.001, partial eta squared = 0.32; see Figure 2A) and temporo-parietal regions (F(1, 91) = 50.78, p < 0.001, partial eta squared = 0.36; see Figure 2B). Notably, FC between the posterior hippocampi and frontal regions was negative for many individuals (see Figure 2). Post-hoc one-sample t-tests were run within each group to determine whether FC scores from the posterior hippocampi to frontal were significantly less than zero. This resulted in significant negative FC for controls from the posterior right hippocampus to left frontal (adjusted p = 0.02), and from the posterior left hippocampus to right frontal (adjusted p = 0.02), and in RTLE, from the posterior left hippocampus to right frontal (adjusted p = 0.03). None were significant in LTLE.

Figure 2.

Resting-state FC patterns from hippocampi to frontal and temporo-parietal regions across groups. A) Anterior and posterior hippocampal FC to frontal regions. B) Anterior and posterior hippocampal FC to temporo-parietal regions. FC from the bilateral anterior hippocampi was significantly stronger than from the bilateral posterior hippocampi to both the frontal and temporo-parietal regions in all three groups (both p < 0.001). C) FC of bilateral anterior hippocampi to frontal regions. D) FC of bilateral anterior hippocampi to temporo-parietal regions. No significant differences were found between groups from the bilateral anterior hippocampi to the frontal regions. In contrast, FC from the bilateral anterior hippocampus to temporo-parietal regions revealed a significant Group x Hippocampus interaction (p = 0.01), such that in the RTLE group, the right (ipsilateral) hippocampus was significantly less connected to temporo-parietal cortex, while in the LTLE group the left (ipsilateral) hippocampus was significantly less connected. Error bars represent 95% confidence intervals.

* p < 0.05

*** p < 0.001

3.1.2. FC of Anterior Hippocampi to Frontal and Temporo-Parietal Regions

No significant differences were found between groups for resting-state FC patterns from the anterior hippocampi to frontal regions (F(2, 91) = 0.18, p = 0.83, partial eta squared = 0.004; see Figure 2C). In contrast to the frontal regions, FC from the anterior hippocampi to temporo-parietal regions revealed a significant relationship that differed by group (F(2, 91) = 4.61, p = 0.01, partial eta squared = 0.09). Specifically, in the RTLE group, the right (ipsilateral) hippocampus (Mean = 2.08 ± 2.15) was significantly less connected to temporo-parietal cortex than the left hippocampus (Mean = 2.87 ± 2.84, p = 0.02), while in the LTLE group the left (ipsilateral) hippocampus (Mean = 1.76 ± 2.59) was significantly less connected to temporo-parietal cortex than the right hippocampus (Mean = 2.79 ± 2.41, p = 0.04; see Figure 2D).

3.1.3. FC of Posterior Hippocampi to Frontal and Temporo-Parietal Regions

No significant differences were found in resting-state FC patterns between groups from the posterior hippocampi to the left or right frontal regions (F(2, 91) = 0.51, p = 0.60, partial eta squared = 0.01). Similarly, no significant differences were found between groups from the posterior hippocampi to the left or right temporo-parietal regions (F(2, 91) = 1.24, p = 0.30, partial eta squared = 0.03).

3.2. Resting-State Laterality Index

LI-Rest measures were found to produce lateralized values in the majority of subjects (68.1%), defined as values ≥ 0.20 for left lateralization and ≤ −0.20 for right lateralization, in accordance with the literature on the most common threshold [46], No significant differences were found in LI-Rest lateralization between groups overall (F(2, 91) = 1.16, p = 0.20, partial eta squared = 0.02), or comparing LI-Rest between frontal regions vs. temporo-parietal regions (F(1, 91) = 1.36, p = 0.25, partial eta squared = 0.02). However, there was a significant Group x Region interaction, in which we found significant differences in LI-Rest lateralization between groups for the temporo-parietal regions, (F(2, 91) = 4.70, p = 0.01, partial eta squared = 0.09), such that the LTLE group showed significantly more right lateralization than RTLE in the temporo-parietal regions (LTLE Mean= −0.24 ± 0.43, RTLE Mean= 0.06 ± 0.52, p = 0.02; see Figure 3B). Across all groups, no significant differences were found in LI-Rest between the anterior vs. posterior hippocampus (F(1, 91) = 2.36, p = 0.13, partial eta squared = 0.03).

Figure 3.

LI-Rest values calculated from both anterior and posterior hippocampal seeds to frontal and temporo-parietal regions. A) No significant differences were found between groups for LI-Rest measured from the hippocampi to frontal regions. B) The LTLE group showed significantly more right lateralization for LI-Rest measured from hippocampi to temporo-parietal regions closest to the seizure focus compared to RTLE patients (LTLE Mean = −0.24 ± 0.43, RTLE Mean = 0.06 ± 0.52; p = 0.02).

Error bars represent standard deviations.

*p < 0.05

3.3. Relationship to Epilepsy Measures

In RTLE, no significant correlations were found between age of onset, duration of disease, monthly seizure frequency, or hippocampal volume and resting-state regional FC or LI-Rest measures (all p > 0.05). In LTLE, there were no significant correlations between duration of disease or monthly seizure frequency and resting-state regional FC or LI-Rest measures (all p > 0.05). However, in LTLE there were two significant correlations between age of onset and resting-state FC from the left anterior hippocampus to left frontal regions (r_s = 0.75, adjusted p = 0.03) and resting-state FC from the left anterior hippocampus to left temporo-parietal regions (r_s = 0.81, adjusted p = 0.01). In addition, in LTLE there was one significant negative correlation between the left posterior hippocampal volume and resting-state FC from the left posterior hippocampus to right frontal regions (r_s = −0.75, adjusted p = 0.02). Figure 4 displays scatter plots for the significant correlations between age of onset and hippocampal volume to resting-state FC in LTLE, as well as boxplots of control values for the same FC seed and target.

Figure 4.

Epilepsy measures correlate with FC from the left hippocampus to frontal and temporo-parietal regions in LTLE patients. Age of onset is associated with FC from the left anterior hippocampus to A) left frontal regions and B) left temporo-parietal regions. C) Left posterior hippocampal volume is associated with FC from the left posterior hippocampus to right frontal regions. Boxplots next to each scatter plot show the control data for the same FC seed and target, while the dashed black line shows the 5^th to 95^th percentile for easy comparison next to the scatterplots. Note that no linear regression line is shown as these are Spearman rho correlations.

*p < 0.05

3.4. Brain-Behavior Correlations

No significant correlations between neuropsychological language scores and resting-state FC or LI-Rest measures were found which also passed the FDR cut-off. However, there was one correlation in LTLE and two correlations in RTLE that were significant at uncorrected p < 0.05. We report them here as an exploratory analysis that should be interpreted with caution and followed up in a larger sample, with scatter plots displayed in Figure 5: (1) In LTLE, negative correlation between the WAIS/WASI Verbal Comprehension Index and resting-state FC from the left posterior hippocampus to right frontal regions (r = −0.66, p = 0.04, uncorrected), (2) In RTLE, positive correlation between the WAIS/WASI Verbal Comprehension Index and resting-state FC from the left anterior hippocampus to right frontal regions (r = 0.41, p = 0.047, uncorrected), (3) In RTLE, negative correlation between the COWAT Word Generation-Letters and resting-state FC from the left anterior hippocampus to left temporo-parietal regions (r = −0.44, p = 0.03, uncorrected).

Figure 5.

Correlations to Neuropsychological Language Measures (Non-Significant). Three Pearson r correlations between resting-state FC and language measures are presented that were significant at p < 0.05, but did not pass the FDR correction. They are presented here to illustrate possible relationships between resting-state FC and language abilities in RTLE (A and B) and LTLE (C), but should be interpreted with caution and followed-up in a larger sample size.

4. Discussion

This study compared resting-state regional FC patterns and lateralization between the bilateral anterior and posterior hippocampi to frontal and temporo-parietal cortex, including language cortex, in RTLE and LTLE patients and healthy controls. We found that resting-state FC from the hippocampi to temporo-parietal cortex closer to the seizure foci, but not the more distal frontal cortex, revealed unique patterns in RTLE and LTLE patients. Specifically, each patient group showed stronger temporo-parietal connectivity from their contralateral anterior hippocampus, while controls showed similar connectivity from both hippocampi to temporo-parietal cortex. Furthermore, we also found that LTLE showed significantly more abnormal right lateralization than RTLE in the temporo-parietal regions, but not frontal regions, as evidenced by laterality measured from resting-state FC — suggesting that hippocampal-temporo-parietal FC measures acquired from resting-state may be a useful tool to detect lateralization in cases where task-LI is not suitable. As a whole, our results are suggestive of abnormalities in the hippocampal networks relevant to language processing, resulting from the presence of temporal lobe epilepsy, and align with previous findings of decreased FC from the epileptic ipsilateral hippocampus to frontal-temporal-parietal networks [17]. Our results show that this decreased FC is driven by reductions in connectivity to the temporo-parietal regions.

Our finding of stronger connectivity from the contralateral hippocampus to temporo-parietal regions in both LTLE and RTLE is suggestive of a compensatory relationship that may underpin language function. At present, the emerging picture of the direct hippocampal role in language appears to be its contribution to rapid processing of lexical-semantic information. Specifically, the evidence suggests that hippocampal theta oscillations are involved in mapping incoming words onto semantic knowledge beyond just word retrieval in service of generating sentence meaning [15,16]. Given that understanding sentence meaning in auditory comprehension is relatively spared in RTLE and LTLE and often on par with healthy controls [6], it’s possible that our results suggest this could be underpinned by a successful compensatory relationship between the non-epileptic contralateral hippocampus to bilateral temporal regions crucial to comprehension, including both anterior and posterior temporal regions, and in particular the left middle temporal gyrus [47]. Thus, it would be interesting for a future study to include measures of auditory comprehension to assess whether greater FC between the ipsilateral hippocampus to contralateral temporal regions is associated with better comprehension abilities.

We also found that hippocampal network resting-state FC was stronger from the anterior compared to the posterior hippocampus to frontal and temporo-parietal regions across all groups, which corroborates previous studies finding greater structural and functional connectivity between temporal lobe areas and the anterior hippocampus [48,49]. However, it was notable that we found evidence of negative FC (or anti-correlations) between the posterior hippocampus and language-related cortex, in particular to the contralateral frontal regions across many subjects. Because the exact cognitive roles of the anterior and posterior hippocampus are still being mapped out [50], it appears this may be a relatively new finding. This could relate to previous work finding anti-correlations in a task-inhibiting network during resting-state that included frontal and parahippocampal regions [51].

The associations we found between resting-state FC and measures of epilepsy including age of onset and left posterior hippocampal volume in LTLE further indicates that the presence of temporal lobe epilepsy impacts hippocampal networks, especially in LTLE patients. The age of onset correlations suggest that patients whose seizures began in early childhood may functionally recruit their ipsilateral (impaired) hippocampus less strongly than later onset patients. The hippocampal volume correlation in LTLE patients suggests that the hippocampal atrophy contributes to the lack of the negative or anti-correlation between left posterior hippocampus and right frontal regions that we observed in controls. These correlations suggest possible evidence of abnormalities (hippocampal atrophy and less anti-correlated FC) and reorganization (weaker FC from ipsilateral hippocampus in earlier onset) resulting from the manifestation of temporal lobe epilepsy that should be studied further. Yet, it was notable that we did not find significant associations between language scores and resting-state FC or LI-Rest. We included these analyses to determine if patterns of resting-state FC would show a relationship with language abilities, but we only found hints of possible associations to measures of language in RTLE and LTLE. Upon reviewing the scatter plots in Figure 5, it’s worth noting that the relationships in RTLE do not appear robust, and would likely not hold in a larger sample size. However, the relationship in LTLE could have been impacted by Type II error (incorrectly accepting the null hypothesis when it’s false) due to a loss of power as a result of the low sample size.

The primary limitation of the present study is the small sample size of temporal lobe epilepsy patients, particularly LTLE patients (n=13). The smaller sample sizes impacted the power to detect significant relationships with neuropsychological language measures, as only a subset of our sample had data on any one test. In addition, our language tests were carried out in the context of a long neuropsychological testing environment, and it’s possible that these tests were not sensitive enough to the types of language abilities that may be impacted by resting-state FC patterns, which could necessitate probing of more language faculties and/or more naturalistic language tests. A second limitation is that our regions of interest were gyral-based, in which all cortex was encompassed on the particular gyrus, extending to the deepest point of the adjacent sulcus on both sides. This meant that sulci were included within the gyral ROIs, but were often split between two regions and may not have included the entire length of the sulci. It’s possible that we would have found functional associations between language abilities and perisylvian cortex along the entirety of the lateral sulcus.

In conclusion, the results of this study add to our understanding of how the presence of temporal lobe epilepsy impacts hippocampal networks relevant for language processing, specifically by potentially decreasing recruitment of the ipsilateral epileptic hippocampus in favor of stronger connectivity from the contralateral (and potentially healthier) hippocampus to nearby temporo-parietal cortex in TLE patients. Our findings suggest the need for follow-up studies to determine the nature of the impact of hippocampal network FC on language abilities. Future work should also aim to study the potential for pre-surgical hippocampal network FC to predict post-surgical language outcomes, as resting-state FC could be an ideal noninvasive measure to assess plasticity and reorganization of these networks over time. Continued research like the present work will be crucial to determine how temporal lobe epilepsy leads to long-lasting language impairments, with the ultimate goal of identifying early language intervention and treatment options for patients.

Highlights.

• The role of the hippocampus in networks related to language is poorly understood in TLE

• LTLE/RTLE showed stronger FC from contralateral hippocampus to temporo-parietal cortex

• LTLE showed greater right lateralization measured from resting-state

• Hippocampal network FC relates to age of onset and left anterior hippocampal volume in LTLE