Announcer The following program features simulated voices generated for educational and philosophical exploration.

Adam Ramirez Good evening. I'm Adam Ramirez.

Jennifer Brooks And I'm Jennifer Brooks. Welcome to Simulectics Radio.

Adam Ramirez Tonight we're examining grid cells—neurons in the medial entorhinal cortex that fire at multiple locations arranged in a periodic hexagonal lattice across the environment. Discovered in 2005, grid cells are thought to provide a metric for space, enabling navigation and spatial memory. The striking hexagonal firing pattern suggests a specific computational mechanism, possibly implementing path integration by updating position estimates based on self-motion signals. However, the question is whether grid cells are fundamental computational units performing coordinate transformations, or whether the hexagonal pattern is an epiphenomenon emerging from network dynamics without direct computational significance.

Jennifer Brooks The grid cell discovery was remarkable because of how regular the spatial firing patterns are. Unlike place cells in hippocampus, which fire at single locations, each grid cell fires at multiple regularly spaced locations forming a triangular lattice. Different grid cells have different spatial scales and orientations, but within a cell, the spacing and geometry are precise. This regularity suggests an underlying computational structure. Path integration—the ability to track position by integrating velocity signals—is a natural candidate computation because it requires a coordinate system. But there's ongoing debate about whether the hexagonal pattern is necessary for path integration or simply one possible solution among many.

Adam Ramirez To explore whether grid cells implement path integration through specific mechanisms or whether their patterns are epiphenomenal, we're joined by Dr. Edvard Moser, a neuroscientist at the Norwegian University of Science and Technology who co-discovered grid cells with May-Britt Moser. His research focuses on neural mechanisms of spatial representation and memory. Dr. Moser, welcome.

Dr. Edvard Moser Thank you. The relationship between grid cell geometry and computational function remains one of the most interesting questions in systems neuroscience.

Jennifer Brooks Let's start with the basics. What exactly did you observe when you first recorded from entorhinal cortex, and what made grid cells so unexpected?

Dr. Edvard Moser We were recording from layer II of medial entorhinal cortex while rats explored open environments. We expected to find cells that encoded spatial information, given entorhinal cortex's anatomical position as the main cortical input to hippocampus. But we didn't expect the striking regularity we observed. Individual neurons fired at multiple locations, and when you mapped out all the firing locations, they formed a perfect hexagonal grid spanning the entire environment. The spacing was consistent—about fifty to seventy centimeters for the coarsest grids in rats—and the pattern was stable across sessions. Different cells had different spatial scales, with grid spacing increasing in discrete modules as you moved dorsoventrally through entorhinal cortex. This systematic organization suggested a fundamental computational architecture.

Adam Ramirez What was the initial hypothesis about what grid cells were computing? Why hexagons specifically?

Dr. Edvard Moser The hexagonal pattern immediately suggested path integration. If you're tracking your position by integrating velocity, you need some coordinate representation. A hexagonal grid is mathematically optimal for covering two-dimensional space with minimal distortion—it's the densest packing of circles in a plane. Early models proposed that grid cells form attractor networks where recurrent connectivity creates stable activity bumps that move in response to velocity inputs. The hexagonal symmetry emerges naturally from certain network architectures with specific connectivity patterns. However, hexagons aren't the only possible solution. You could implement path integration with other grid geometries or even without grids at all. So the question became whether the hexagonal pattern is necessary for the computation or merely a byproduct of the specific circuit architecture evolution produced.

Jennifer Brooks What's the evidence that grid cells actually perform path integration rather than simply reflecting it?

Dr. Edvard Moser Several lines of evidence support a causal role. First, grid cells update their firing patterns based on self-motion even in darkness, without external sensory cues. The animal moves, and the grid pattern shifts accordingly, suggesting internal position tracking. Second, when you manipulate the animal's perception of self-motion—for instance, by rotating visual cues during movement—grid patterns respond in ways consistent with integrating the manipulated motion signals. Third, lesions of medial entorhinal cortex impair path integration behavior. However, the strongest evidence would be showing that disrupting grid cell activity specifically impairs path integration while leaving other navigation strategies intact. Some optogenetic studies have approached this, but cleanly separating grid cell function from other entorhinal cortex contributions remains challenging.

Adam Ramirez How do grid cells receive velocity information? What are the inputs that drive grid firing?

Dr. Edvard Moser This is still being worked out. Grid cells receive inputs from head direction cells, which encode the animal's directional heading, and likely from neurons encoding running speed. These could provide the velocity components needed for path integration. There are also inputs from place cells in hippocampus, though the relationship is complicated because place cells also receive inputs from grid cells, creating a loop. Recent work has identified neurons in medial entorhinal cortex that encode specific combinations of speed and direction, which could serve as velocity signals. But we don't yet have a complete circuit diagram showing how velocity information is transformed into the spatial updating of grid patterns. The computational mechanisms likely involve complex interactions between multiple cell types and may differ across the dorsoventral axis where grid scales vary.

Jennifer Brooks You mentioned attractor models. How do these explain the hexagonal pattern mechanistically?

Dr. Edvard Moser Continuous attractor models propose that grid cells form a network where activity is confined to a localized bump that can move smoothly through neural state space. The position of the bump represents the animal's position in the environment. Recurrent connections are structured such that nearby neurons in grid space excite each other while more distant neurons inhibit each other. This creates a landscape of stable states. The hexagonal pattern emerges when you set up the connectivity to have threefold rotational symmetry—neurons are connected to others at specific angles. When velocity inputs push the activity bump, it moves through the network, and the periodic connectivity causes the bump to reappear at regularly spaced locations, creating the grid pattern. However, these models make strong assumptions about connectivity that haven't been fully verified anatomically.

Adam Ramirez Are there alternative models that don't rely on continuous attractors?

Dr. Edvard Moser Yes. One class of models proposes that grid patterns emerge from interference between oscillations at different frequencies. If you have neurons that oscillate at slightly different frequencies as a function of running direction, the interference between these oscillations can create periodic spatial firing patterns. These models don't require fine-tuned recurrent connectivity—the grids emerge from feedforward processing of oscillatory inputs. Another approach involves learning-based models where grid-like patterns emerge through synaptic plasticity driven by experience. These models don't assume grids are hardwired but instead show that grid patterns can develop from more generic learning rules applied to spatial navigation tasks. Each model class has strengths and weaknesses, and determining which best matches biological reality requires detailed comparison with experimental data.

Jennifer Brooks What does the developmental evidence show? Are grid cells present from birth, or do they develop through experience?

Dr. Edvard Moser Grid cells emerge during development. In young rats, before extensive exploration experience, neurons in medial entorhinal cortex show spatial tuning but not the characteristic hexagonal periodicity of adult grids. As animals explore their environment, grid patterns gradually become more regular and stable. This suggests that experience plays a role in refining grid structure, consistent with learning-based models. However, we don't yet know whether the basic capacity for grid-like firing is innate and just needs calibration, or whether the entire pattern is learned from scratch. There's also evidence that grid cell modules mature at different rates, with finer-scale grids developing later. Understanding the developmental trajectory and what experience is necessary could help distinguish between model classes.

Adam Ramirez How do grid cells relate to place cells in hippocampus? Is one representation more fundamental?

Dr. Edvard Moser This is a crucial question. Place cells were discovered first, in the 1970s, and for decades were considered the neural basis of spatial representation. Grid cells were later understood as providing input to place cells. Current models suggest that place cell firing—which is highly location-specific—could emerge from summing inputs from multiple grid cells with different spatial phases. Because grid cells tile space periodically, their combined input is unique at each location, giving place cells their specificity. However, information also flows from hippocampus back to entorhinal cortex, so the relationship is bidirectional. Some evidence suggests grid cells can function independently of hippocampus, maintaining stable grids even when hippocampal activity is disrupted. But place cells require grid cell input for their spatial specificity. This suggests grid cells may be more fundamental for metric spatial representation, while place cells create location-specific maps useful for memory.

Jennifer Brooks Do grid cells only encode physical space, or have they been found to represent other variables?

Dr. Edvard Moser This is an active research area. Recent work has found grid-like representations in non-spatial domains. For instance, when animals are trained to navigate abstract task spaces—like sequences of sensory stimuli or conceptual variables—neural activity in entorhinal cortex shows periodic structure similar to spatial grids. This suggests grid cells might implement a more general computational principle of metric representation, not limited to physical space. However, whether these non-spatial grids use the same cellular mechanisms as spatial grids, or whether they're distinct populations with different computational properties, remains unclear. If grid cells are truly domain-general metric systems, it would suggest the hexagonal pattern reflects a fundamental solution to representing continuous variables, strengthening the case that it's computationally meaningful rather than epiphenomenal.

Adam Ramirez What happens to grid cells when the environment changes—when you reshape the arena or move the animal to a new room?

Dr. Edvard Moser Grid cells show interesting remapping properties. When you move an animal to a completely new environment, grid patterns can rotate and shift, but they maintain their hexagonal structure and spacing. This is called global remapping. The grid is preserved, but its alignment to external landmarks changes. Within a familiar environment, if you subtly manipulate cues, you see partial remapping where some grid properties adjust while others remain stable. This flexibility suggests grid cells aren't rigidly tied to specific external features but rather maintain an internal metric that can be anchored to different reference frames. However, severe distortions of the environment—like stretching or compressing space—can disrupt grid regularity, suggesting the system has limits. Understanding how grid cells balance stability and flexibility is important for understanding their computational role.

Jennifer Brooks Are there pathological conditions or genetic manipulations that specifically disrupt grid cells while leaving other spatial representations intact?

Dr. Edvard Moser This is technically challenging, but there's progress. Some studies have used optogenetics to disrupt medial entorhinal cortex activity during specific task phases and found impairments in path integration but not in other navigation strategies like following visual cues. Genetic manipulations targeting specific ion channels or receptors expressed in grid cells can alter grid spacing or regularity. For instance, disrupting HCN channels affects grid scale, suggesting these channels contribute to setting the spatial frequency of firing. However, cleanly separating grid cells from other entorhinal cortex cell types—like border cells, head direction cells, and speed cells—is difficult because they're intermixed. Ideally, we'd have genetic markers that uniquely identify grid cells, allowing cell-type-specific manipulations. Such tools are being developed and should provide clearer causal evidence for grid cell function.

Adam Ramirez How well do computational models of grid cells match real neural data? What discrepancies remain?

Dr. Edvard Moser Models capture many features well—the hexagonal geometry, modular organization, and integration of velocity signals. However, discrepancies exist. Real grid cells show more variability than models predict. Grid fields aren't perfectly symmetric, and spacing isn't perfectly uniform. Models struggle to explain how grid cells respond to environmental boundaries and geometric features. Some models predict that grids should be more fragile to noise than they actually are. The relationship between grid cell activity and theta rhythm oscillations, which are prominent during navigation, isn't fully captured by all models. And we don't yet have models that comprehensively explain the full diversity of entorhinal cortex cell types and how they interact. Bridging the gap between idealized models and messy biological reality is an ongoing challenge.

Jennifer Brooks Do grid cells exist in all mammals? What about other vertebrates or even invertebrates?

Dr. Edvard Moser Grid cells have been found in multiple mammalian species—rats, mice, bats, and recently in humans using intracranial recordings. The basic hexagonal pattern is conserved, though there are species-specific variations in scale and organization. In bats, which navigate three-dimensional space, there's evidence for three-dimensional grid-like representations, though the geometry is complex and debated. Whether non-mammalian vertebrates or invertebrates have grid cells is an open question. Birds navigate using different neural structures, and insects perform sophisticated navigation using much smaller brains. If grid cells reflect a universal computational principle, we might expect functionally analogous systems across species, even if the specific neural implementation differs. But if grids are specific to mammalian entorhinal cortex architecture, they might be one solution among many that evolution has produced for spatial navigation.

Adam Ramirez Looking forward, what experiments would definitively test whether the hexagonal pattern is computationally necessary?

Dr. Edvard Moser We need experiments that independently manipulate grid geometry and path integration performance. If you could disrupt the hexagonal regularity—perhaps through targeted optogenetic perturbations or genetic manipulations—while preserving overall entorhinal cortex function, and test whether path integration is specifically impaired, that would be strong evidence. Another approach is developing artificial systems that implement different geometric codes and comparing their performance on navigation tasks to see whether hexagonal grids offer advantages. We could also examine whether grid cells in non-spatial domains show the same geometric properties, which would support domain-general computational significance. Finally, detailed circuit mapping combined with computational modeling could reveal whether the connectivity patterns necessary for hexagonal grids are actually present in the biological tissue, testing attractor model predictions.

Jennifer Brooks Dr. Moser, thank you for clarifying what grid cells reveal about spatial computation and what remains uncertain about their functional significance.

Dr. Edvard Moser Thank you. Grid cells provide a fascinating window into how the brain represents continuous variables, but whether their geometry is computational necessity or architectural accident is still being determined.

Adam Ramirez That's our program. Until tomorrow, stay critical.

Jennifer Brooks And keep questioning. Good night.