Prediction and Prevention of Verbal Memory Decline after Temporal Lobectomy

Commentary

Resective surgery is currently the treatment of choice for intractable temporal lobe epilepsy (TLE), achieving long-term seizure-freedom in up to 50 to 60 percent of patients, even without a clear MRI lesion (1). This seizure control may come at the expense of various functional deficits, including an upper quadrant visual field cut, naming difficulties, and a side-dependent memory impairment, with verbal memory at highest risk following dominant mesial temporal resections and visual memory with nondominant temporal lobe surgery. In fact, “significant” memory decline occurs in 30 to 60 percent of patients after dominant anterior temporal lobectomy (2), and in 33 to 50 percent after left selective amygdalohippocampectomy (3). It is therefore essential to develop and validate tools to accurately predict both the occurrence and severity of postoperative memory change to better inform the presurgical counseling and decision making. At another, deeper level, it behooves the epilepsy community to identify reliable strategies that may prevent or reduce this risk—not solely predict it. Knowledge gained from the two articles chosen for this commentary advances both domains: The prediction and prevention of memory decline following epilepsy surgery.

The first article by Sidhu et al. joins a growing body of literature in evolving our appreciation of memory as a “network function” while simultaneously highlighting the critical role of the mesial temporal structures. In a study of 44 patients with unilateral mesial TLE due to hippocampal sclerosis, the authors found extensive bilateral extratemporal and lateral temporal reorganization of fMRI activation during memory encoding. However, better memory function was mainly dependent on reorganization within the medial temporal lobe itself (particularly in the posterior hippocampus), with additional key activation of the orbitofrontal cortex, anterior cingulum, and insula. The second paper by Bonelli et al. further reinforces the importance of this medial temporal reorganization by showing that a robust preoperative activation of the left posterior hippocampus during word encoding can reliably predict better verbal memory outcomes after temporal lobectomy. So, what do these findings mean in relation to our two domains of interest: prediction and prevention of postoperative memory decline?

The question of memory outcome prediction is indeed worth considering because traditional predictors such as preoperative memory performance on neuropsychological testing, age at epilepsy onset, and even WADA testing are insufficient to provide a comprehensive assessment of surgical memory risk (2). fMRI naturally emerges then as a noninvasive additional investigative method. Theoretically, a tool that measures memory function should help assess memory reserve and organization. The issue is that practically, many intricacies need to be addressed before a memory predictive fMRI becomes widely accepted as a clinically useful tool.

First, the ideal “memory measuring” protocol needs to be defined. Studies that used scene-encoding tasks activating the mesial temporal lobes bilaterally—possibly masking deficits due to a unilateral dysfunction (4, 5)—were less predictive of postoperative verbal memory decline than those using more specific paradigms, such as word list learning, story recall, and delayed recall, as was done in the manuscript at hand by Bonelli et al. and others (6, 7). Besides the activation paradigms, fMRI findings vary depending on several technical definitions, including the delineation of the seed region within the mesial temporal structures, MRI acquisition parameters, and comparative control population (8). These multiple parameters may account for variable results in the current literature, and underscore our second point: the “ideal” memory fMRI protocol—once determined—needs to be reproducible and validated in the clinical setting. One challenge includes the transition between “group data” and “group analyses” to clinically meaningful information at the individual patient level. If we consider the current findings by Bonelli et al., for example, it makes intuitive sense that patients whose verbal memory function “lives” in the anterior hippocampus—typically resected in TLE surgery due to hippocampal sclerosis—would be at higher risk for memory decline than patients whose verbal memory “lives” in the typically spared posterior hippocampus and tail: The challenge is that for an individual patient considering a temporal lobectomy and needing to assess his or her risk for memory decline with surgery, a verbal memory paradigm will likely activate the whole hippocampus-anterior and posterior to varying degrees, and an estimation of this particular patient's risk in relation to the degree and pattern of his or her hippocampal activation will be helpful. Lastly, given current evidence that a language lateralizing fMRI may provide the dual advantage of assessing both language and verbal memory risks with temporal lobe resections (2), it would be interesting to determine the additional predictive contribution of a “memory fMRI.”

Beyond predicting a postoperative memory decline, the significance of critical work presented in the manuscripts at hand extends to indicating potential preventative measures. One such obvious measure proposed here and elsewhere is to limit the extent of hippocampal resection whenever possible. This again makes excellent intuitive sense, as the goal of any epilepsy surgery, by definition, is to remove the smallest amount of brain tissue required to achieve seizure freedom. Several challenging questions remain: First, the tail and posterior hippocampus are already often spared in a “standard temporal lobectomy”; in fact, sparing them in patients with “inefficient” preoperative reorganization of verbal memory (i.e., one relying more on anterior hippocampus) would not necessarily be expected to improve memory outcomes. Conversely, it is difficult to predict what the residual verbal memory function would actually be if the anterior hippocampus is “spared” but effectively disconnected because the remaining sclerosed and epileptic hippocampus needed to be removed. Second, it remains to be seen how the fMRI episodic memory maps generated in cohorts with hippocampal sclerosis translate to patients with nonlesional TLE, who are actually at the highest risk of verbal memory decline after surgery. The last untapped opportunity for prevention resides in a better understanding of the mechanisms underlying various patterns of presurgical memory reorganization in patients with epilepsy. Data presented here by Sidhu et al. and by other investigators (8) suggest that there are multiple potential patterns of altered resting functional connectivity and task-driven fMRI activation in patients with otherwise clinically similar mesial TLE, albeit with some patterns being more efficient than others and correlating with better baseline and postoperative memory function. Multiple factors may account for this variability, including variability in learning strategies during development (9) or genetic predisposition (10). It is unclear whether preoperative dysfunctional networks can be modified prior to surgery to optimize a patient's chance of a successful outcome.

While several questions remain, thoughtful work such as the studies highlighted here help in moving us closer to answers.