BOLDly Going Where Few Researchers Have Gone Before—Leveraging Language-Related Hippocampal Activations to Predict Postoperative Memory Decline

Commentary

Identifying accurate and reliable biomarkers for predicting postoperative memory decline is a holy grail for epilepsy specialists in the surgical setting. Although functional MRI has risen as an attractive and noninvasive modality for lateralizing language profiles, there remains disagreement as to whether fMRI, with its technical limitations and task challenges, can reliably capture memory-related activations and inform risk for postoperative decline. Given these barriers, Binder et al ¹ advocated for the use of fMRI language lateralization as a surrogate marker for predicting verbal memory decline. This approach is based on biological models that strongly link the verbal episodic memory encoding system with the language system. According to this model, peri-sylvian language lateralization should be a reliable indicator of medial temporal verbal memory lateralization, and thus, a strong predictor of verbal memory decline. However, this approach only works if the model of language/memory colateralization is accurate—a model which is still under investigation and may not hold in the face of neuropathology.

Guided by a similar model proposing language and declarative memory as a functional unit, ² Salleles et al ³ test whether hippocampal activations observed during language fMRI could be used to enhance or refine memory fMRI in the prediction of postoperative memory decline. In a cohort of 46 consecutive patients with temporal lobe epilepsy (TLE) with available language fMRI, memory fMRI, and comprehensive neuropsychological testing, the research team addressed 2 main questions: (1) What is the relative sensitivity of memory fMRI versus language fMRI for eliciting hippocampal activations? and (2) Are hippocampal activations observed during language versus memory fMRI superior for predicting postoperative memory decline?

The authors found evidence for the usefulness of both memory and language fMRI hippocampal activations. Through a series of classification analyses, the authors found that memory fMRI was more sensitive than language fMRI for eliciting hippocampal activations across patients (95% for memory vs 65% for language). However, when hippocampal activations were detected with language fMRI, it resulted in superior performance compared to memory fMRI for predicting both global and nonverbal memory outcomes. Conversely, memory fMRI still emerged as a stronger predictor of verbal memory decline in the full sample, regardless of hippocampal activations. Of interest, in the patients who showed postoperative worsening (arguably the most instructive group), memory fMRI accurately predicted global memory deterioration in 64%, whereas language fMRI predicted a slightly lower percentage (55%). The authors’ final analysis involved examining the predictive power of using both techniques together, which they report yielded the highest prediction accuracy across patients with left and right seizure onsets. The authors conclude that whereas memory fMRI elicited greater hippocampal activations than language fMRI, hippocampal activations during language tasks have high specificity, and could help to refine memory fMRI predictions in the presurgical evaluation.

This study raises at least 2 intriguing questions for which researchers and clinicians remain divided. From a theoretical standpoint, what do hippocampal activations reflect during language processing and how are they related to memory? And second, for clinicians who use fMRI in the context of presurgical planning, is it necessary to test both modalities or could using only one modality be enough to infer lateralization of the other?

Regarding the first question on hippocampal activations, there is increasing support for a direct role of the hippocampus in lexicosemantic aspects of language, beyond just episodic memory. Strong hippocampal fMRI activations have been observed during the processing of lexical knowledge in healthy controls ⁴ as well as in children and adults with focal epilepsy. ⁵ Recent evidence from intracranial recordings has also revealed an association between interictal epileptiform discharges in the hippocampus and word finding difficulty. ⁶ Furthermore, clinical studies demonstrate that hippocampal pathology in TLE influences language lateralization, ⁷ intrahemispheric language reorganization, ⁸ and pre- and postoperative naming decline. ⁹ Although the exact role of the hippocampus in language is debated, intriguing new data from human neurophysiology suggests that the same hippocampal theta oscillations that support memory also subserve online language usage. Given the hippocampus’ role in relational binding, it may play a direct role in language by continuously relating incoming words to stored semantic knowledge. ¹⁰ Collectively, these studies align with theories espousing the interdependency of language and memory and advocate for a shared language and memory network.

However, can we always rely on the concept of a shared and colateralized network in the presence of epilepsy-related pathology? Although the presence of hippocampal sclerosis may increase the likelihood of both memory and language reorganization, ⁸ this is not guaranteed. In the current study, even though the majority had hippocampal sclerosis, full or partial concordance in hippocampal activations was only observed in 72% of patients. This leaves us with a 1/3 of patients with completely discordant fMRI profiles, which could reflect fully segregated language and memory networks. Importantly, when complete discordance was observed in the subset of patients who underwent surgery, neither task emerged as a clear winner for predicting postoperative memory decline. Although this analysis was based on only 5 patients, it should give clinicians pause in terms of assuming superiority of one task over the other in the most challenging of patients.

Strengths of this study include the analysis of both language and memory fMRI, the comprehensive neuropsychological data available at pre- and postsurgery, and the level of detail provided on individual patients in the cohort. Study limitations include categorical rather than continuous designations of concordance/discordance and memory outcomes, the qualitative nature of the fMRI judgments, and the relatively small sample size of patients who were included in the postoperative analyses. Although the qualitative nature of the analysis limits power, this approach has appeal in that it emulates current clinical practices at most centers, which rely on visual inspection of images and categorical decisions. Finally, the authors acknowledge that a major limitation is an inherent selection bias. That is, 9 of the patients whose fMRI findings suggested a high risk for memory worsening were not operated on in the first place. Although not uncommon, this bias limits their memory outcome results to patients with relatively low risk profiles.

This study adds to a growing body of literature challenging the narrow view of the hippocampus as a “memory-specific” structure and advances the notion of a shared neural mechanism, or at the very least an overlapping of regions, for memory and language that may hold for a majority of patients. In line with prior practice guidelines ¹¹ and a recent meta-analysis, ¹² memory fMRI may be a somewhat stronger predictor of verbal memory decline than language fMRI. However, questions abound as to how to evaluate risk for memory decline in up to 1/3 of patients with discordant findings where neither task excelled. In these cases, relying more heavily on clinical approaches (eg, nomograms) ¹³ or other neurobiological indicators (eg, extra-hippocampal activations or white matter changes) to define risk may be prudent and help to maximize successful memory outcomes.