Localization of Function

In brief

Localization of function is the working assumption beneath almost all brain anatomy and brain imaging: that the brain is divided into parts, and that particular mental capacities, speech, face recognition, fear, the movement of the right hand, are carried out by particular structures. Ask why a stroke can rob a person of language while leaving memory intact, or what an fMRI scanner is looking for when it watches "which region lights up," and the answer rests on this one idea. It was born disreputable, as the skull-reading of phrenology, then rescued by nineteenth-century clinicians who matched specific deficits to specific damage. Today it is the default frame of neuroscience. It is also under steady attack from a rival picture in which the important unit is not the region but the network, and in which asking where a function "is" can be the wrong question.

The full treatment

The problem it answers

For most of history the brain was treated as a single organ with a single job, or as the seat of a soul that was not divisible at all. But behavior is not uniform. Damage to the head does not degrade the mind evenly, like turning down a dimmer; it removes some capacities and spares others in strange, specific patterns. A person may lose the ability to speak yet understand everything said to them, or recognize objects but not faces, or move but not feel. Localization is the hypothesis that explains this: the mind is built from parts, those parts are physically separate, and injury to a part removes its function while leaving the rest running. This is the first principle beneath every clinical brain exam and every imaging study, because both are attempts to map function onto place.

The phrenological false start

The first systematic version was wrong in almost every particular but right in its core intuition. Franz Joseph Gall (1758 to 1828), a Viennese anatomist, proposed that the brain is an assembly of some two dozen "organs," each responsible for a mental faculty such as language, benevolence, or acquisitiveness, and that an organ used heavily would grow larger, pushing out the skull above it. Read the bumps on a head, phrenologists claimed, and you could read the character within. The faculties were made up, the skull does not track the brain's surface, and the method became a Victorian parlor trick and, worse, a prop for racial ranking. Yet Gall had insisted on two ideas that survived him: that the cortex is the seat of mind, and that different parts do different things.

Phrenology's downfall came from the laboratory. Pierre Flourens (1794 to 1867) removed parts of animal brains and found that recovery was often broad and that many losses did not respect Gall's tidy map, evidence for a brain that acted more as a whole. For a time the pendulum swung hard against localization.

Broca, Wernicke, and the real evidence

It swung back on the strength of two patients. In 1861 the French surgeon Paul Broca (1824 to 1880) examined a man, Louis Victor Leborgne, known as "Tan" because that syllable was nearly all he could say, though he understood others and was otherwise lucid. When Leborgne died, Broca found a lesion in the left frontal lobe. More cases followed the same pattern, and the region, the left inferior frontal gyrus, became Broca's area, the first mental function tied to a specific piece of cortex on solid post-mortem evidence. Broca also noticed that the damage was on the left in case after case, the founding observation of hemispheric specialization that split-brain-research would later push much further (see split-brain-research).

In 1874 the German physician Carl Wernicke (1848 to 1905) described a different aphasia: patients who spoke fluently, even copiously, but in nonsense, and who could not understand speech. Their damage lay farther back, in the left superior temporal region, now Wernicke's area. Two language deficits, two locations, two jobs. Wernicke's model, that comprehension and production sit in separate connected regions, made localization a research program rather than a slogan. Korbinian Brodmann (1868 to 1918) then mapped the cortex into numbered areas by their microscopic cell structure in 1909, giving the field a coordinate system still cited today.

From lesions to imaging

For a century the evidence was clinical: you learned what a region did by finding people in whom it had failed, as with Broca's patients, the frontal-lobe injury of Phineas Gage, or the amnesic patient H.M. (see phineas-gage-and-h-m). Functional magnetic resonance imaging, developed after Seiji Ogawa's discovery in 1990 that blood oxygenation changes the MRI signal, promised to skip the injury. fMRI measures the BOLD signal, a proxy for blood flow that rises where neurons are more active. Put a healthy person in the scanner, have them read or fear or decide, subtract a control condition, and the leftover activity shows, in principle, where that function lives. Localization moved from the autopsy table to the living, working brain, and the results filled journals with colored maps.

Lineage

Localization descends from the faculty psychology of Gall, purged of its errors, and from the clinical neurology of Broca and Wernicke, who supplied the discipline it lacked: match a precise deficit to precise damage. Its microanatomy comes from Brodmann. It is one instance of a deeper move in psychology, explaining the mind by decomposing it into parts (see levels-of-explanation), the same instinct behind the working-memory model and modular theories of cognition. Its great internal opponent is the network tradition, running from Karl Lashley's early challenge through modern connectomics, which shares the goal of explaining behavior biologically but denies that single regions are the right unit.

The strongest case for it

The evidence for some localization is overwhelming and not seriously in dispute. Primary sensory and motor functions are mapped so reliably that neurosurgeons stimulate the cortex during awake surgery and can predict which finger will twitch. Destroy the primary visual cortex and the person goes cortically blind while the eyes work perfectly. The double dissociation, the logical backbone of the whole enterprise, is powerful: if damage to region A knocks out function X but spares Y, and damage to region B knocks out Y but spares X, then X and Y are separable and separately housed. Broca's and Wernicke's aphasias form exactly such a pair. Localization also earns its keep clinically every day: it lets a physician infer the site of a stroke from the pattern of loss, and lets a surgeon plan around an area whose removal would cost a patient their speech. A theory that guides the knife has passed a stringent test.

The strongest case against it

The critique is not that localization is false but that it is routinely overstated, and that for higher functions the region may be the wrong unit entirely.

The earliest serious objection came from Karl Lashley (1890 to 1958). Searching for the "engram," the physical trace of a memory, he trained rats and then removed pieces of cortex, expecting to find the spot where the memory sat. He never did. What predicted the loss was how much cortex he removed, not where, findings he summarized as mass action and equipotentiality: the cortex acts as a whole, and one part can take over for another. Memory, at least, seemed not to be localized in Broca's sense.

The modern objection targets fMRI directly. Russell Poldrack named its central logical flaw in 2006: the reverse-inference problem. A scan can show that a region is more active during a task. It cannot, by itself, show that the task engages the function the region is famous for. If the amygdala lights up, it is tempting to conclude the subject felt fear, but the amygdala responds to novelty, salience, and much else, so the inference runs backward and often fails. Related is the charge of "blobology": studies that produce a colored blob on a brain slice and treat the location as an explanation, when naming where something happens explains nothing about how. In 2009 Edward Vul and colleagues showed that a swath of highly cited social-neuroscience findings rested on a circular analysis, selecting voxels by the very correlation then reported, inflating the numbers to impossible heights, a paper that circulated under the title "voodoo correlations." The same year, Craig Bennett and colleagues ran a standard fMRI analysis on a dead Atlantic salmon and found "activation," a deliberate demonstration that without proper correction for multiple comparisons the method manufactures signal from noise.

The deepest critique is conceptual. On the network view, associated with Marcus Raichle's discovery of the default-mode network and Marsel Mesulam's account of large-scale distributed systems, complex functions such as attention, memory, and language are not performed by any single area but by the dynamic interaction of many, and the same region participates in different functions depending on the network it is coupled with at the moment. Asking where language "is" can then be as confused as asking which part of an orchestra the symphony is in.

Where it stands now

The field has settled into a synthesis rather than a winner. Strict, one-region-one-function localization holds cleanly for primary sensory and motor areas and for a handful of remarkably specific higher regions, such as the fusiform face area. For most higher cognition, the consensus has shifted toward networks: functions are localized to distributed circuits, not points, and the honest map is one of nodes and connections rather than labeled blobs. The methodological reckoning was real. Standards for fMRI, correction for multiple comparisons, pre-registration, larger samples, tightened sharply after 2009, part of psychology's broader replication crisis (see the-replication-crisis). Localization survives, then, as it was always destined to: narrowed, qualified, and far more careful about the difference between finding where activity occurs and explaining what the brain is doing.

Test yourself

The next time you read that scientists have "found the brain region for" love, lying, or political belief, ask two questions the field learned the hard way. First, does that region do only that, or does it also light up for a dozen other things, in which case its activation tells you little? Second, is the claim that the function happens in that spot, or merely that the spot is one node in a wider network? The gap between those two readings is the whole modern history of this idea.