Abstracts

Rationale:
Episodic memory impairment is common in temporal lobe epilepsy (TLE) because of the importance of mesial temporal structures to memory (Gleissner et al., Epilepsia 2002; 43: 87-95). Abnormal functional connectivity (FC) and structural volume (gray matter; GM) have both been associated with memory deficits in temporal lobe epilepsy (TLE), (Doucet et al., PLoS One 2016; 11: e0154660).  Few studies have examined the degree to which these two sets of memory-related abnormities are concordant with regard to the regions implicated in the memory deficit.  It is also unclear which of these abnormalities is more important to episodic memory performance in TLE. Accordingly, in this project, we seek to address two questions:  1) Do the areas that show abnormal FC during episodic memory task performance also show abnormal GM volume? ; 2) Does episodic memory performance depend on the integrity of both FC and GM volume, and which of these is a better predictor of memory performance. Our working hypothesis is that concordant integrity in both function and structure will best serve successful memory, with abnormalities in either sufficient to cause memory failure.  As TLE patients may implement memory with a reorganized network, we will examine these questions from the perspective of brain responses in both the ictal and non-ictal hemisphere.
Method:
Thirty-two subjects with left TLE (n=14, Good recall = 60% recall accuracy or better, Subsequent Memory Effect,  SME; n=18, Poor recall = lower than 60% SME) and 24 age-matched healthy controls (HC, n=14, Good; n=10, Poor), were participants in this study. The fMRI scan was obtained on a 3T Philips MR Scanner using a paired-associate verbal memory (PAM) task. Data preprocessing was done using fMRIPrep 1.5.8 (Esteban O, et al., Nat Methods 2019; 16: 111-16). FC for the PAM task was calculated through Generalized Psychophysiologic analysis (gPPI) using the CONN toolbox v17.f (McLaren et al., Neuroimage 2012; 61: 1277-86). Regional gray matter (GM) volumetric analysis was carried out the using CAT12 toolbox (Farokhian et al., Front Neurol 2017; 8: 428). Structural and FC network effects on memory were examined through a Support Vector Machine learning algorithm.
Results:
Utilizing the ictal hippocampus as seed, we determined the FC differences between Good and Poor PAM performance as a function of group status (LTLE vs. HC); interaction term). Relative to the HCs, Good performance in the LTLE group was associated with strong FC involving 9 regions (see Figure 1a), bilaterally. The same gPPI analysis utilizing the non-ictal hippocampus as seed revealed Good performance in the LTLE group compared to HCs was associated with strong FC to 8 regions, mostly right-temporal in nature (see Figure 1b). Next, we tested for LTLE/HC differences in the GM volume of the regions found significant in the above gPPI analyses (17 regions). Of these 17, 10 areas showed significantly lower GM volume in LTLE (p< .05). Lastly, within the LTLE group, we trained a linear SVM regression model with the above subset of significant FCs (beta weights) and the matching GM regions (volumes) serving as predictors of PAM accuracy (response variable). The SVM model was significant [RMSE= 0.18; R2 = 0.63, MSE= 0.03, MAE= 0.14], indicating that both FCs and the underlying GM volumes reliably contribute to PAM accuracy.
Conclusion:
The results suggest that when FC and underlying GM structure show concordant abnormality relative to matched HCs strong predictive relationships with episodic memory performance are observed, with stronger concordance reflective of poorer performance.
Funding:
:
Funding:
: J.T., NIH/NINDS, R01 NS112816-01.