Two Sides of the Same Coin: The Hippocampus as a Common Neural Substrate for Model-Based Planning and Spatial Memory

Oliver M. Vikbladh, Michael R. Meager, John A. King, Karen Blackmon, Orrin Devinsky, Daphna Shohamy, Neil Burgess, Nathaniel D. DawbioRxiv 2018 · DOI

TL;DR

This study investigates whether the hippocampus, known for spatial memory, also plays a crucial role in model-based planning. By testing epilepsy patients with hippocampal damage and healthy controls on both tasks, they found that hippocampal damage impairs both functions and that these functions are linked in healthy individuals, suggesting a shared neural substrate.

What Problem Does This Paper Address?

Humans and animals can plan complex actions by mentally simulating future outcomes, a process called "model-based planning." While the hippocampus is well-established as critical for spatial memory (creating "cognitive maps"), its role in model-based planning is less clear. Emerging theories suggest these two functions might rely on the same hippocampal mechanisms, but direct evidence linking hippocampal damage to deficits in both planning and spatial memory, and showing their interdependence, has been limited. This paper aims to fill that gap by directly testing this hypothesis.

Methods

The researchers employed two main behavioral tasks:

  1. Spatial Memory Task: Participants navigated a virtual arena and had to remember object locations. This task was designed to distinguish between "place memory" (allocentric, relative to the environment) and "response memory" (egocentric, relative to a landmark).
  2. Sequential-Decision Making Task (Two-step Markov Decision Task): Participants made choices in a game-like scenario to maximize rewards. This task was designed to measure reliance on "model-based planning" (using a mental model of the task to predict outcomes) versus "model-free learning" (learning from direct experience and reward).

The study involved two groups:

  • Patients: 19 individuals with intractable epilepsy who underwent unilateral anterior temporal lobectomy (ATL) with hippocampal resection.
  • Healthy Controls: 19 age- and IQ-matched individuals.

To analyze the data, they used:

  • Linear Mixed-Effects Regression: To model performance on the spatial memory task, accounting for individual differences and repeated measures.
  • Binomial Mixed-Effects Logistic Regression: To model choices in the decision-making task, capturing the strategy (model-based vs. model-free).
  • Full Computational Learning Model: A more complex model was also used to verify findings and capture trial-by-trial dependencies.

Crucially, they analyzed lesion sizes (overall and hippocampal) to correlate with performance deficits.

Datasets

Not explicitly discussed as "datasets" in the traditional machine learning sense. The study utilized behavioral data collected from human participants (19 epilepsy patients and 19 healthy controls) performing two specific cognitive tasks.

Key Insights and Ideas

  • Hippocampus as a Dual-Purpose Brain Region: The most significant finding is the strong evidence suggesting the hippocampus is not just for spatial navigation but also for abstract planning. This challenges a strict separation of functions within this brain area.
  • Planning and Spatial Memory are Intertwined: In healthy individuals, performance on the spatial memory task and the model-based planning task were correlated. This suggests they draw on common cognitive resources, likely within the hippocampus.
  • Damage to the Right Hippocampus is Particularly Detrimental: The study found that damage to the right hippocampus was more strongly associated with deficits in both planning and, to a lesser extent, spatial memory. This hints at lateralization of these functions.
  • Patients Compensate with Model-Free Learning: When model-based planning was impaired due to hippocampal damage, patients showed a compensatory shift towards using simpler, habitual "model-free" strategies. This highlights the brain's ability to adapt, albeit with a less sophisticated approach.

Key Results

  • Impaired Performance in Patients: Epilepsy patients with ATL and hippocampal resection showed significantly worse performance on both the spatial memory task (specifically place memory) and the model-based planning task compared to healthy controls.
  • Correlation in Controls, Not Patients: In healthy controls, performance on the spatial memory task was significantly correlated with the degree of model-based planning. This correlation was absent in patients, suggesting that hippocampal damage disrupts this link.
  • Lesion Size Matters: The size of the hippocampal lesion, particularly on the right side, was significantly related to the severity of the model-based planning deficits.
  • Right Hippocampus Dominance: The effects of hippocampal damage were more pronounced for lesions on the right side, suggesting a lateralized role for the right hippocampus in both planning and spatial memory.

Limitations and Future Work

  • Causality and Specificity: While the study shows strong correlations and deficits following lesions, definitively proving causality for specific hippocampal subregions is challenging due to the nature of surgical lesions affecting multiple temporal lobe structures. The authors acknowledge that ATL lesions affect more than just the hippocampus.
  • Chronic Epilepsy: The potential influence of chronic epilepsy on cognitive function, independent of the surgical intervention, is a noted caveat.
  • Task Abstraction vs. Spatial Interpretation: The authors mention that the "cover story" of the planning task (rocket trips to planets) might have elicited a spatial interpretation, potentially blurring the lines between abstract and spatial planning.
  • Future Work: The authors suggest future research should focus on studying hippocampal activity in tasks that manipulate environmental relationships to understand how this knowledge is leveraged for decision-making. Further work is needed to disentangle the roles of different hippocampal subregions and surrounding temporal lobe structures.

Summary updated

Abstract

Abstract Little is known about the neural mechanisms that allow humans and animals to plan actions using knowledge of task contingencies. Emerging theories hypothesize that it involves the same hippocampal mechanisms that support self-localization and memory for locations. Yet, there is limited direct evidence for the link between model-based planning and the hippocampal place map. We addressed this by investigating model-based planning and place memory in healthy controls and epilepsy patients treated using unilateral anterior temporal lobectomy with hippocampal resection. We found that both functions were impaired in the patient group. Specifically, the planning impairment was related to right hippocampal lesion size, controlling for overall lesion size. Furthermore, planning and place memory covaried with one another, but only in neurologically intact controls, consistent with both functions relying on the same structure in the healthy brain. These findings clarify the scope of hippocampal contributions to behavior and the neural mechanism of model-based planning.
BibTeX
@misc{vikbladh2018two,
  author = {Oliver M. Vikbladh and Michael R. Meager and John A. King and Karen Blackmon and Orrin Devinsky and Daphna Shohamy and Neil Burgess and Nathaniel D. Daw},
  title = {{Two Sides of the Same Coin: The Hippocampus as a Common Neural Substrate for Model-Based Planning and Spatial Memory}},
  year = {2018},
  doi = {10.1101/359232},
  howpublished = {bioRxiv},
}
Figure