What Makes the Human Brain Big

A chimpanzee’s brain weighs roughly a third as much as yours. Yet both are built from the same cell types, arranged in the same six cortical layers, by the same kinds of progenitor.

So what is actually different? You have spent several lessons on how the cortex is made: radial glia lining the ventricle, intermediate progenitors just above them, neurons climbing to an inside-out stack of layers. This lesson asks which steps in that process changed in our lineage, and how anyone has tested it. The candidates are more progenitors, a different kind of progenitor, a handful of genes that only humans carry, and more time. None of them, on its own, made us human.

How big, measured how: a scaled-up primate brain

Start with the number everyone quotes: about 86 billion neurons. It comes from one study. Frederico Azevedo, Suzana Herculano-Houzel and colleagues (J Comp Neurol, 2009) took the brains of four adult men, dissolved each region into a suspension of cell nuclei, and counted the nuclei stained for NeuN, a protein found in neuronal nuclei. The method is called the isotropic fractionator. The mean was 86.1 billion neurons, with a standard deviation of about 8 billion, and about 85 billion other cells. So the figure is an average of four men, with a real spread, and it says nothing yet about women, children or variation across people.

Where those neurons sit is surprising. About 69 billion are in the cerebellum, and only about 16 billion are in the cerebral cortex, even though the cortex makes up most of the brain’s mass. Herculano-Houzel’s comparisons across species led her to describe the human brain as a primate brain scaled up: it holds about the number of neurons you would predict for a primate brain of its size. Primate brains pack neurons more densely than rodent brains do, so a rodent brain with as many neurons would be far larger.

What does stand out is the cortex as a sheet. Across mammals, cortical thickness changes little while area changes enormously. The human cortex is only about twice as thick as a mouse’s, but its surface area is roughly a thousand times larger (figures summarized by Pasko Rakic, Nat Rev Neurosci, 2009). Much of that sheet is folded to fit inside the skull. So the question becomes: how does a developing brain make more area without making a thicker wall?

More columns, not a thicker cortex

Recall from that lesson that radial glia first divide symmetrically, each making two radial glia, and later divide asymmetrically, each making one radial glial cell and one neuron or intermediate progenitor. Professor Deppmann teaches three modes of neurogenesis: direct, where a radial glial cell makes a neuron; indirect, where it makes an intermediate progenitor that divides once into two neurons; and a third route, through outer radial glia. Keep those three in mind.

Rakic, working in the fetal rhesus monkey, had shown that neurons born at one spot in the ventricular zone climb the same radial glial fibre and stack into a column, with the birth order setting their layer. In 1988 (Science) he proposed the radial unit hypothesis. Each patch of progenitors, a proliferative unit, makes one ontogenetic column. The number of units sets the area of the cortex. The number of neurons each unit makes sets its thickness.

Here is the arithmetic that makes the idea powerful. While founders divide symmetrically, the pool doubles each round. Once they switch to asymmetric division, each extra division adds only one neuron to a column.

units after k symmetric rounds = N0 × 2^k; neurons per unit grow by about one per extra asymmetric division

Doubling is exponential and adding is linear. In 1995 (Trends Neurosci) Rakic proposed that a delay of only a few cell cycles before the switch to asymmetric division could account for a large increase in cortical area across primates. One test came from mice. Anjen Chenn and Christopher Walsh (Science, 2002) made transgenic mice whose neural progenitors expressed a stabilized form of β-catenin, which kept more of them re-entering the cell cycle. The mice grew enlarged brains with a greater cortical surface area, and the normally smooth cortex buckled into folds.

Design problem

Suppose each extra round of symmetric division in the founder pool doubles the number of radial units, and each unit still makes the same number of neurons. If one species has three more rounds than another, how much larger is its cortical surface area, and what happens to cortical thickness?

A stem cell without a ventricular address

The radial unit hypothesis assumes that all progenitors live at the ventricle. In primates that turned out to be incomplete. In fetal monkey cortex, Iain Smart, Colette Dehay, Henry Kennedy and colleagues (Cereb Cortex, 2002) described a subventricular zone split into an inner and an outer part, the outer subventricular zone. What was dividing there?

In 2010, David Hansen, Jan Lui, Philip Parker and Arnold Kriegstein (Nature) labelled cells in living slices of human fetal cortex and filmed them. In the outer subventricular zone they found cells that expressed PAX6 and SOX2, the transcription factors of radial glia, and kept a long basal fibre toward the brain surface. Unlike classical radial glia, they had no process touching the ventricle. Before dividing, a cell moved its cell body rapidly along its fibre, and then went into mitosis there. Daughters followed in the films could inherit the fibre and divide again, or become intermediate progenitors expressing TBR2. These outer radial glia, in other words, renewed themselves and fed the indirect route of neurogenesis from a site far from the ventricle.

The same group found that many of these cells expressed HES1, a Notch target, and that DAPT, a drug that blocks γ-secretase, shifted cells in cultured slices from the SOX2-high state toward the TBR2-high state. Because DAPT blocks other γ-secretase targets too, this points to a role for Notch signalling without proving that Notch alone is responsible.

Outer radial glia are not uniquely human. In the same year Simone Fietz and colleagues in Wieland Huttner’s laboratory described related progenitors in human and in ferret cortex (Nat Neurosci, 2010). A few similar cells exist in the embryonic mouse (Wang et al., Nat Neurosci, 2011; Shitamukai et al., J Neurosci, 2011). The difference is one of degree: these cells are abundant in primates and ferret and rare in mouse, and the human outer subventricular zone is especially large. Because their fibres diverge as they climb, they fan the radial scaffold outward, which favours growth in area and has been linked to folding in the ferret. The marmoset, a monkey with a nearly smooth cortex, also has an outer subventricular zone, so these cells are not sufficient for folding on their own.

Genes that only humans have

Gene duplication is a common way for a lineage to gain a new regulator. Three duplicated genes whose functional copies are found in humans and not in chimpanzees have been tested for effects on the cortex. Note the system each was tested in.

ARHGAP11B. Marta Florio, Huttner and colleagues (Science, 2015) found this partial copy of ARHGAP11A expressed in human cortical progenitors. Electroporated into the embryonic mouse cortex, it increased the number of basal progenitors, especially basal radial glia. Later work from the same laboratory expressed it in ferret embryos and in marmoset fetuses (Heide et al., Science, 2020), where it enlarged the neocortex and increased upper-layer neurons.

NOTCH2NL. Two groups published together in Cell in 2018. Ian Fiddes and colleagues in David Haussler’s group traced these human-specific genes, which derive from part of NOTCH2, and found that deleting them from human stem cells made cortical organoids whose progenitors differentiated early. Ikuo Suzuki and colleagues in Pierre Vanderhaeghen’s laboratory found that NOTCH2NL raises Notch signalling and, expressed in embryonic mouse cortex, kept progenitors dividing and delayed neuron production. The genes sit at 1q21.1, a region where deletions and duplications are associated with smaller and larger heads (that lesson).

SRGAP2C. Cécile Charrier, Franck Polleux and colleagues (Cell, 2012) showed that this truncated copy of SRGAP2 blocks the function of the ancestral protein. Expressed in mouse cortical neurons, it slowed the maturation of dendritic spines and increased their density in the adult. It acts on neurons rather than progenitors.

Each result comes from adding or removing a human gene in a mouse, ferret, marmoset or organoid. None was tested by removing the gene from a developing human brain, and none shows that one gene made the human cortex.

Taking longer: development on a slower clock

The last difference is time. Cortical neurogenesis takes about a week in the mouse and roughly two months in the rhesus monkey (Rakic, Science, 1974), and lasts longer still in humans. Every extra week of neurogenesis gives founders and outer radial glia more cycles to run.

Synapses follow the same slow clock. Peter Huttenlocher and Arun Dabholkar (J Comp Neurol, 1997) counted synapses in human cortex and found that synaptic density in prefrontal cortex peaks later, and declines later, than in auditory cortex, with pruning continuing into adolescence. Zdravko Petanjek, Rakic and colleagues (PNAS, 2011) found that spine density on pyramidal neurons of human prefrontal cortex stays above adult levels into the third decade of life. Keeping juvenile features for longer is called neoteny. SRGAP2C is one candidate for it at the level of single synapses.

Is the slow clock set by the body or by the cells? When the Vanderhaeghen laboratory transplanted human cortical neurons made from stem cells into the cortex of newborn mice (Linaro et al., Neuron, 2019), the human neurons matured over many months, close to a human time course, while the mouse neurons around them had long since matured. At least part of the timing travels with the human cells.