Hippocampus-associated causal network of structural covariance measuring structural damage progression in temporal lobe epilepsy

Abstract

In mesial temporal lobe epilepsy (mTLE), the causal relationship of morphometric alterations between hippocampus and the other regions, that is, how the hippocampal atrophy leads to progressive morphometric alterations in the epileptic network regions remains largely unclear. In this study, a causal network of structural covariance (CaSCN) was proposed to map the causal effects of hippocampal atrophy on the network-based morphometric alterations in mTLE. It was hypothesized that if cross-sectional morphometric MRI data could be attributed temporal information, for example, by sequencing the data according to disease progression information, GCA would be a feasible approach for constructing a CaSCN. Based on a large cohort of mTLE patients (n = 108), the hippocampus-associated CaSCN revealed that the hippocampus and the thalamus were prominent nodes exerting causal effects (i.e., GM reduction) on other regions and that the prefrontal cortex and cerebellum were prominent nodes being subject to causal effects. Intriguingly, compensatory increased gray matter volume in the contralateral temporal region and post cingulate cortex were also detected. The method unraveled richer information for mapping network atrophy in mTLE relative to the traditional methods of stage-specific comparisons and structured covariance network. This study provided new evidence on the network spread mechanism in terms of the causal influence of hippocampal atrophy on progressive brain structural alterations in mTLE. Hum Brain Mapp 38:753-766, 2017. © 2016 Wiley Periodicals, Inc.

Keywords: granger causality analysis; morphometric MRI; network of structural covariance; progression; temporal lobe epilepsy.

Figure 1

Gray matter atrophy and its relationships with progressive factors in mTLE. A: Group comparison of GMV between mTLE and healthy controls using two‐sample t‐test (P < 0.01, FDR correction). Decreased GMV in mTLE was distributed at the mesial and lateral temporal lobes ipsilateral to the epileptogenic side, the bilateral thalamus, frontal lobes and cerebellum. GMV increase was found in the controlateral amygdale and posterior cingulate cortex by lowering threshold (P < 0.05, FDR correction). B: Correlation analyses between GMV and progressive factors of epilepsy duration (Upper) and estimated number of lifetime seizures (Lower) in mTLE. For epilepsy durations, negative correlation was found in the ipsilateral hippocampus, bilateral frontal lobe, and cerebellar hemispheres; for number of lifetime seizures, negative correlation was found in the ipsilateral thalamus and bilateral caudate nuclei in addition to the correlation results of epilepsy duration. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 2

Progressive patterns of stage‐specific GMV alterations in mTLE relative to healthy controls. A: Stages categorized by epilepsy duration (stage I/II/III/IV = 0.5–4.5 years/4.5–11.6 years/11.6–16.5 years/16.6–40years). With increase of stages, GMV reductions progressively expand from mesial temporal structure to the thalamus, frontal lobe, and cerebellum. B: Stages categorized by estimated number of lifetime seizures (stage I/II/III/IV = 5–60 times/60–300 times/300–800 times/>800 times). With increase of stages, GMV reductions progressively expand from mesial temporal structure to the lateral temporal lobe, thalamus, frontal lobe, caudate heads, and cerebellum. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 3

Patterns of hippocampus‐associated structural covariance networks (SCN). A: SCN of mTLE, B: SCN of the healthy controls. Based on a large cohort of cross‐sectional morphometric data of patients and controls, SCNs were constructed seeding at the hippocampus showing significant GMV reduction in mTLE. In each group of patients and HCs, the averaged GMV values in the seeding region were extracted from each subject and used as a regressor in the General‐linear‐model to produce VBM‐SCNs of each group. C: Comparison of SCN between patients with mTLE and healthy controls. Comparing analysis was performed using multi‐regression model‐based linear‐interaction analysis. In the patients with mTLE, increased synchronization of GMV alterations (GMV covariance) with seeding region (hippocampus) was presented in the ipsilateral mesial temporal regions, and decreased synchronization was presented in the ipsilateral lateral temporal cortex and contralateral hippocampus. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 4

Hippocampus‐associated causal networks of structural covariance (CaSCNs) in mTLE. CaSCNs were constructed by applying granger causal (GC) analysis to a large‐cohort of sequenced morphometric data according to progressive factors of epilepsy duration (A) and estimated number of lifetime seizures (B). Seeding region was identical to that in SCN analysis. In both CaSCNs with epilepsy duration and number of lifetime seizures, the lateral temporal lobe ipsilateral to epileptogenic focus, bilateral lateral prefrontal cortices, medial prefrontal cortex and thalamus present consistent positive GC value, the contralateral temporal cortex and posterior cingulate cortex present consistent negative GC value. The contralateral amygdala shows negative GC value only in the CaSCN with epilepsy duration, and the bilateral insula show positive GC value only in the CaSCN with estimated number of lifetime. The bilateral basal ganglia including the caudate head and putamen show opposite GC values in these two CaSCNs. The positive GC value denotes that the GMV reduction in the region has causal relationship with, and is preceded by hippocampal atrophy, which may imply seizure damaging effect from hippocampus. The negative GC value denotes that the regions show an opposite (enlarged) GMV alteration caused by hippocampal atrophy, which is explained as compensatory effect of brain structure. The lower figures present causal relationship between hippocampal atrophy and GMV alterations in the other regions within a glass brain. Time series consist of averaged GMV values extracted from the regions showing GC connectivity with hippocampus. X‐axis: Progressive factors (patients sequenced by progressive factors), Y‐axis: GMV values. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 5

ROI‐based analysis of causal networks of structural covariance. Five ROIs were selected from the results of overall GMV atrophy in two‐sample t‐test group comparisons. In both networks based on data sequencing with epilepsy duration (A) and number of lifetime seizures (B), in addition to the hippocampus, the thalamus was another prominent node exerting causal effects on other regions, and the prefrontal cortices and the cerebellum were prominent regions being subject to causal effects from other regions. [Color figure can be viewed at http://wileyonlinelibrary.com]