Lecture 8: Navigation I

What this lecture covers

The lecture opens by resolving the Haxby debate from the previous session (what kinds of causal evidence answer the claim that "selective" regions secretly carry information about other categories), then shifts to a new topic: navigation. It surveys remarkable animal navigation feats, breaks navigation down into component problems, and uses the parahippocampal place area (PPA) as a concrete case study for how researchers narrow down what a brain region represents through a sequence of controlled experiments. It closes by introducing fMRI adaptation as a method that can detect neural information even when multi-voxel pattern analysis (MVPA) fails.

After watching, you can list the component problems of navigation (recognizing a place, sensing spatial layout, heading direction, route planning, reorientation), describe how researchers used image manipulations to test what drives the PPA's response, and explain how fMRI adaptation differs from MVPA and why it is sometimes needed.

Key ideas

• Causal evidence beyond pattern information: TMS, direct intracranial stimulation, and prosopagnosia (loss of face perception with intact object perception) all argue that, despite information about non-face objects existing in the pattern of FFA responses, only face perception is causally affected when the region is disrupted, suggesting the extra pattern information may be epiphenomenal.
• Two fundamental navigation questions: "where am I?" (recognizing a specific place, knowing what kind of place it is, sensing spatial layout, knowing exits) and "how do I get from here to there?" (route planning around obstacles, using heading direction, and beaconing toward a visible or audible goal).
• Cognitive maps: Tolman's classic rat experiments showed animals learn an abstracted spatial representation of an environment, not just a fixed sequence of turns, since rats took a novel direct route to a goal after their trained route was blocked.
• The parahippocampal place area (PPA): a region that responds much more strongly to scenes than to objects; systematic tests (removing objects, using empty rooms, using familiar vs. unfamiliar scenes) point to it representing spatial layout rather than object content, familiarity, or scene meaning.
• Ruling out the rectilinearity confound: a challenge argued that the PPA's scene selectivity might really be a preference for rectilinear shapes; a 2x2 design (scene/face by rectilinear/curvilinear) and a pattern-analysis check showed rectilinearity does not account for scene selectivity in real scenes.
• Multiple scene-selective regions: besides the PPA, retrosplenial cortex (RSC), the occipital place area (OPA, formerly TOS), and the hippocampus form a broader navigation system, each with different apparent functions.
• A functional double dissociation: TMS over OPA impaired discrimination of scene structure but not faces, while TMS over the occipital face area impaired face discrimination but not scenes, giving strong evidence the two regions serve different functions.
• fMRI adaptation: a method that presents pairs of stimuli back-to-back and measures whether the response is lower for repeated ("same") stimuli than for different ones; unlike MVPA, it can reveal discriminative neural information even when the relevant neurons are spatially intermixed within a voxel.

Walkthrough

Resolving the Haxby debate (0:10)

The lecture reviews Haxby's challenge (that low-response regions can still carry pattern information about other categories) and discusses what kind of evidence answers it: TMS effects specific to faces, intracranial stimulation producing face percepts (not other percepts), and prosopagnosia patients who lose face recognition but keep other object recognition. Together these suggest the extra pattern information detected by Haxby-style analysis may not be causally used by the brain, and the lecture explicitly flags this as an open, debated question.

Extending Haxby's method to new questions (7:14)

The lecture shows how the same within-category-vs-between-category correlation logic used for the FFA can test whether the PPA discriminates beach scenes from city scenes, generalizing Haxby's method into a general tool for probing what information a region holds, while cautioning that detectable information doesn't guarantee the brain uses it.

Animal navigation feats and the structure of the problem (13:20)

Examples include monarch butterfly multi-generational migration, loggerhead turtles returning to their birth beach after 20 years, bats navigating in the dark, and desert ants (Cataglyphis) computing a direct vector home without landmarks. From these, the lecture defines navigation's two core questions (where am I, and how do I get from here to there) and their sub-components: recognizing a specific place, judging what kind of place it is, sensing spatial layout, knowing heading direction, planning routes around obstacles, and reorienting when lost. Tolman's rat maze experiment is used to introduce the concept of a cognitive map.

Discovering and characterizing the PPA (29:28)

The lecture recounts how the PPA was found somewhat by accident when scanning subjects looking at scenes versus objects, then walks through a series of controlled comparisons: scenes with objects removed (leaving empty rooms) still evoke a strong response, ruling out object content or complexity as the driver; scenes broken into scrambled pieces reduce the response, ruling out low-level visual coverage; and familiar (MIT) versus unfamiliar (Tufts) scenes, counterbalanced across student groups, produce similar responses, ruling out recognition of the specific place.

Testing and rejecting the rectilinearity confound (44:33)

A competing account argued the PPA actually prefers rectilinear shapes, which happen to be common in scenes. A 2x2 design crossing category (scene/face) with shape (rectilinear/curvilinear) found only a main effect of category, no interaction with rectilinearity. A pattern-based reanalysis (following the same Haxby logic) confirmed the PPA's activity pattern carries strong category information but not rectilinearity information.

Causal evidence from intracranial stimulation (54:44)

A rare case of direct electrical stimulation in a patient's PPA is described: stimulating the region produced vivid, spontaneous percepts of specific remembered places (a train station, a closet), while stimulating adjacent non-scene-selective sites produced nothing, offering causal evidence that PPA activity is linked to scene perception.

The broader scene-processing network (1:01:52)

Beyond the PPA, the lecture introduces retrosplenial cortex (RSC), the occipital place area (OPA), and the hippocampus. RSC responds more to familiar places and appears involved in tasks requiring knowledge of location and heading, consistent with a patient whose RSC damage left him able to recognize places but unable to judge directions from them. A TMS study shows a double dissociation: disrupting OPA impairs scene discrimination but not face discrimination, while disrupting the occipital face area shows the opposite pattern.

Introducing fMRI adaptation (1:15:00)

The lecture explains a key limitation of MVPA: it can only detect information when the relevant neural populations are spatially clustered enough to differ across voxels; if information-carrying neurons are spatially intermixed, MVPA can miss it entirely. fMRI adaptation addresses this by presenting stimulus pairs back-to-back and measuring whether repeated ("same") stimuli produce a weaker response than differing ("different") stimuli, revealing discrimination at the neural level regardless of spatial clustering.

Before you watch

• Review the previous lecture's coverage of Haxby's paper and multi-voxel pattern analysis, since this lecture begins by extending that discussion.
• Be familiar with main effects, interactions, and 2x2 experimental designs, which are used again here for the rectilinearity test.
• Recall the concept of a double dissociation from earlier in the course, since it reappears with the OPA/occipital face area TMS result.

Check your understanding

1. What three kinds of causal evidence were used to argue that FFA pattern information about non-face objects may not be behaviorally used, even though it is statistically detectable?
2. How did removing objects from scene images and comparing familiar to unfamiliar scenes help narrow down what drives the PPA's response?
3. What is a cognitive map, and what evidence from Tolman's rat experiment supports its existence?
4. Why doesn't the absence of a main effect of rectilinearity in the PPA's average response rule out the region carrying rectilinearity information in its response pattern?
5. Why can fMRI adaptation detect neural discrimination that MVPA sometimes cannot?