Recent studies have explored the possibility of using preoperative fMRI to predict episodic memory impairment after ATL resection. Given the prominent role played by the hippocampus and other MTL structures in episodic memory, most of this research has focused on preoperative activation in MTL structures. In the first study to examine verbal memory outcome, Richardson and colleagues identified a focus in the anterior hippocampus where asymmetry of activation on fMRI predicted verbal memory outcome on a standardized word list learning test after left ATL resection (Richardson et al., 2004
). Greater activation in this region on the left side relative to the right side predicted greater decline. Subsequent studies by the same authors showed correlations between verbal memory change and preoperative activation in either the left hippocampus (Powell et al., 2008
) or both the left and right hippocampus (Richardson et al., 2006
). Frings et al. showed a weak correlation (1-tailed p
= .077) between postoperative verbal memory change and preoperative fMRI activation asymmetry in the hippocampus (Frings et al., 2008
). While encouraging, these studies used very small sample sizes (7–12 left ATL surgery patients) and should therefore be viewed as preliminary. Moreover, the results were not entirely consistent across studies, with two reporting activation asymmetry
as a significant predictor (Richardson et al., 2004
; Frings et al., 2008
) and two reporting unilateral activation on the left side as the main predictor (Richardson et al., 2006
; Powell et al., 2008
). In the studies by Richardson and colleagues, the exact location of the hippocampal ROI that best predicts outcome has also varied from study to study. Finally, a study by Rabin et al. failed to find any relationship between preoperative MTL activation (or activation asymmetry) and postoperative verbal memory outcome (Rabin et al., 2004
One difficulty in using hippocampal activation as a prognostic marker is the relatively small size of this structure, which contains only around a hundred fMRI voxels even when scanning at relatively high resolution. Compared to the entire cerebral hemisphere, which contains many thousands of voxels, measuring activation in the hippocampus thus depends on a much smaller sample size and is consequently more susceptible to random noise, subject motion, and measurement error arising from imperfect identification of hippocampal boundaries. Loss of MRI signal due to macroscopic field inhomogeneity ("signal dropout") often affects the amygdala and occasionally the anterior hippocampus (Constable et al., 2000
; Fransson et al., 2001
; Morawetz et al., 2008
) and represents another technical difficulty for fMRI of the hippocampus. For these reasons, we explored the use of language lateralization as a prognostic marker for verbal memory outcome. The obvious advantage of such an approach is that the ROI over which lateralization is measured can be much larger, potentially offering a more stable and reliable measurement. In addition to this technical advantage, however, there are theoretical reasons to consider language lateralization as a prognostic marker. Evidence suggests that verbal and nonverbal information is encoded in somewhat distinct episodic memory networks, and that these typically have different patterns of hemispheric lateralization (Glosser et al., 1995
; Kelley et al., 1998
; Chiaravalloti & Glosser, 2001
; Golby et al., 2001
; Reber et al., 2002
; Powell et al., 2005
). Furthermore, it seems likely that this specialization for verbal or nonverbal material is not a property of the MTL itself, but is rather a function of the type of information received by the MTL from the ipsilateral hemisphere. If this model is correct, then the MTL in the language-dominant hemisphere is likely to be more critical for supporting verbal episodic memory, and language lateralization should be correlated with verbal memory lateralization.
Analysis of the current patient subgroup is consistent with our previous larger study (Binder et al., 2008
) in showing a correlation between verbal memory change and preoperative language LI. Effect sizes were similar in the two analyses, though the correlations with Delayed Recall outcome did not reach standard significance levels in the subgroup analysis due to the smaller sample size. In the previous study, both fMRI language LI (r = −.432, p
<.001) and Wada language asymmetry (r = −.398, p
<.01) were significant predictors of verbal memory change, lending strong support to the hypothesis that verbal memory lateralization is correlated with language lateralization. In contrast, a material-nonspecific Wada memory asymmetry score (recognition of visually-encoded real objects) was only weakly related to verbal memory outcome (r = −.331, p
In contrast to the correlations with language LI, we found no evidence of a correlation between verbal memory outcome and preoperative hippocampal activation. There were no significant correlations with either activation asymmetry or unilateral activation in the left hippocampus. These null findings held for ROIs targeting the anterior hippocampus, posterior hippocampus, and whole hippocampus.
Notably, these null results cannot be explained by poor data quality or lack of functional activation. All patients in the study had significantly activated voxels detected in the hippocampus. More importantly, the hippocampal LIs, especially the anterior hippocampus LI, showed clear evidence of functional lateralization away from the side of the seizure focus, as expected in chronic temporal lobe epilepsy. Finally, the anterior and whole hippocampus LIs were significantly correlated with Wada memory asymmetry. Thus, this fMRI protocol based on a visual scene encoding task detected hippocampal functional asymmetry preoperatively, but failed to predict postoperative memory outcome.
We believe this apparent paradox can be explained by activation of both verbal and nonverbal episodic memory networks during the scene encoding task. The resulting activation reflects overall asymmetry of the (verbal + nonverbal) episodic memory system and is therefore a good indicator of lateralized MTL damage. However, because the activation represents a mix of verbal and nonverbal encoding processes, it is not a specific indicator of verbal memory lateralization and thus cannot accurately predict verbal memory outcome. In some patients, for example, overall episodic memory is relatively lateralized to the right side while verbal memory remains exclusively on the left, resulting in unexpected verbal memory decline despite preoperative lateralization of activation to the right (). In contrast, language lateralization is predictive of verbal memory outcome because it is closely tied to lateralization of verbal episodic memory processes.
According to this model, predicting verbal memory outcome requires an fMRI contrast that distinguishes verbal from nonverbal memory processes. The work by Richardson and colleagues, cited above, has followed this approach (Richardson et al., 2004
; Richardson et al., 2006
; Powell et al., 2008
). Their patients performed a semantic decision task (e.g., "Is it living or nonliving?") on a series of words during fMRI and were later given a recognition test including the same words. Activation related specifically to verbal encoding was identified by contrasting the subsequently recognized (and therefore successfully encoded) words against words that were not subsequently recognized. The authors then searched for voxels in the MTL where activation or activation asymmetry was correlated with verbal memory change across the patient sample. A limitation of this approach is the initial uncertainty about the locations of these correlated voxels, which have varied somewhat from study to study. In order for the test to be useful in a newly encountered individual, some way of identifying these critical voxels a priori
would seem to be necessary.
The language activation protocol used in the current study also uses a semantic decision task that engages verbal episodic memory encoding. Although the contrast does not focus on successful vs. unsuccessful encoding, the comparison between a semantic word task and a sensory tone task constitutes a "depth of processing" manipulation that is known to modulate verbal episodic memory encoding processes (Craik & Lockhart, 1972
; Otten et al., 2001
; Bartha et al., 2003
). Thus, the critical difference between our approach and the one adopted by Richardson and colleagues is the region of interest over which activation is analyzed. Given the assumption that lateralization of verbal episodic memory processes is tightly coupled to language lateralization, using the larger language network to compute a lateralization index provides a more stable measurement than a small hippocampal ROI and obviates the need to find the set of hippocampus voxels that are predictive of outcome.
In summary, this study shows that preoperative hippocampal activation asymmetry elicited by a scene encoding task is not predictive of verbal memory outcome. Risk of verbal memory decline is likely to be related to lateralization of material-specific verbal memory networks, which are more closely correlated with language lateralization than with overall episodic memory asymmetry.