Epilepsy Did it Again: Memory and Model-Based Planning Studied in Patients After Epilepsy Surgery

Commentary

Over 80 years ago during surgery in patients with epilepsy, Wilder Penfield determined by electrical stimulation the distribution of primary sensory modalities and motor control over the neocortex. ¹ The sensory and motor homunculi were born and indeed developed further. ² Similarly, focal seizures in Broca’s area added to the localization of language expression ³ and even distinct brain expression of each language in bilingual patients. ⁴ Many neurobiological principles about the role of the hippocampus in memory formation in humans originated in the studies of HM, the famous patient in whom bilateral hippocampi were removed for intractable epilepsy by Dr Scoville and the follow-up neuropsychological evaluations performed by Dr Milner. ⁵ To be fair to other conditions, similar memory problems were seen after medial temporal lobe removal (including the hippocampi) in patients with psychosis. ⁶ However, the many flavors of cognition governed by the hippocampus in humans, while being thoroughly investigated, are still unknown. Here, surgical resections as treatment for intractable epilepsy can still bring invaluable insights.

The authors of the present study asked a seminal question: Is the hippocampus involved in model-based planned tasks? It was hypothesized that in humans, planning the tasks (modeling) may utilize similar mechanisms as hippocampal self-localization and memory for location using allocentric approach based on the remote cues. This approach is represented by positioning in a relative coordinate system using “cognitive maps.” On the other hand, model-free learning (habitual response-based behavior) involves dopamine and basal ganglia similar to egocentric stimulus response strategies (such as following directions based on turns). ⁷

The authors used with advantage a pool of patients (n = 19) with anterior temporal lobectomy for intractable epilepsy. This cohort was quite balanced for right and left lobectomy, and within the side of surgery, there were no significant differences in age or sex. Healthy controls (n = 19) were recruited matching the patients for age and sex. The authors also took advantage of advanced imaging that mapped structural brain scans of their patients to either Montreal Neurological Institute template or Harvard-Oxford Brain lexicon. Using this approach, the volume of the lesion (Montreal) or the volume of ablated hippocampus (Harvard) was determined. While the volume of the lesion was larger for patients with right-sided lobectomy, the percentage of the resected hippocampus was relatively comparable on both sides (∼63% on the right, vs ∼54% on the left). This is important information as the volume of the lesion or the hippocampus removed may be significant in correlation with either deficit severity or change in the task performance. The first experiment determined the proportion of model-based versus model-free decisions based on the performance in the 2-step Markov decision task. ⁸ There were 2 levels of decision in this computer-based task. First, the participants decided between 2 spaceships of distinct color that took them with fixed and opposite probabilities (0.7 and 0.3) to 2 different planets. On those planets, they were able to select one of the 2 different aliens specific for each planet. Here each alien was associated with a unique but slowly drifting probability of receiving a monetary reward. In this scenario, model-free decision is based on selecting the spaceship previously leading to a reward. On the other hand, in a model-based decision, the model of reward is created, and even if the reward occurs after an event, which is rare in the model, in the next trial the proband will attempt to avoid the same spaceship. Indeed, a comparison between controls and patients indicated a significant shift from model-based behavior to model-free decisions in patients after temporal lobectomy. In the second experiment based on a virtual reality, participants had to determine the location of 4 objects (4 blocks with different object locations, for each 16 trials) from memory (navigational task). ⁹ Two objects were fixed to the cues in the boundaries of the area (distant cues) and 2 to the landmarks within the area (proximal cues). They found that patients with temporal lobectomy do not commit more errors using the proximal (landmark) cues compared to control participants but they displayed greater errors than controls in positioning the objects relying on the boundary (distal cues). These findings suggest that anterior temporal lobe structures (such as the hippocampus) preferentially use distant cues for navigation (spatial memory). Using statistical models, the authors examined the relationship between model-based planning (first experiment) and distant cue navigation (second experiment). Regression determined a significant correlation for these 2 variables in control participants, indicating that these cognitive approaches are likely correlated. However, this correlation was not present in the patients. Statistics also revealed that these differences were more driven by the right-sided lobe lesions. Interestingly, hippocampal lesion volume on the right was significantly related to shift from model-based to model-free strategy in the Markov decision task, but not for the navigation using distant cues.

The study brought up human investigations close to the information available form rodent studies at least in term of the spatial navigation: The hippocampus is more sensitive for use of distant than proximal navigational cues. Additionally, there are dual navigational systems involving not only the hippocampus but also the basal ganglia. ¹⁰ There are a few limitations to the study. First, the sample size is small (not allowing for robust stratifications) yet this is understandable and despite that the authors should be commended for the incredible study. Second, use of the landmark (a proximal cue) in conjunction with distal cues somewhat dilutes its role as this landmark is always related to the distant cues. The only solution would be a design of 2 different experiments, one based solely on distant cues and one using just a proximal landmark. A wonderful example of spatial, hippocampus-related navigation based on distant cues occurred in a very recent study. ¹¹ In that study, the authors utilized virtual reality navigation in portable devices (smartphones) to simulate an (fun-packed) analogy of Morris Water Maze for humans to distinguish the role of APOE4 allele on hippocampal function. Sensitivity of their test was tremendous even in low numbers of APOE4 probands and they also collected over 10 000 reference individuals for comparison, in whom they found age and sex dependence of the performance in spatial navigation. Third, the authors also acknowledge that their cover story for Markov decision task (spaceships) may have spatial context and therefore may contain hippocampal confounder. Finally, there might have been confounders in terms of additional tissue damage due to epilepsy ongoing until the surgery, an effect which cannot be easily controlled for. There was also resection of additional structures besides the hippocampus with a potential role for navigational and decision tasks. Despite these limitations, this study signifies that model-based planning and spatial (or place) memory operate using the same or very similar anterior temporal lobe substrates, very likely localized in the hippocampus.