Synapses in the Brain: Adhesion, Scaffolds and Astrocytes

A kidney cell has no business at a synapse. Yet if it makes one neuronal protein, neuroligin, the axons that touch it in a dish gather synaptic vesicles at the point of contact and start releasing them, as if they had found a partner.

That result, from Peter Scheiffele and Tito Serafini’s laboratory in 2000, sets the questions for this lesson. What tells a growth cone to become a nerve terminal? What builds the receiving side? What do astrocytes add? And why can a protein that builds synapses in a dish turn out not to be needed for synapses in the animal?

The idea: both partners arrive almost ready

The professor lists three “challenges of synapse formation”: making selective connections, differentiating a nerve terminal, and building a postsynaptic apparatus. Partner choice is the subject of that lesson. This lesson takes the second and third jobs in the brain and spinal cord, the central nervous system.

Start with his comparison, “NMJ vs. Central synapses.” At a mature mammalian neuromuscular junction (see that lesson), a muscle fiber is usually contacted by one motor axon at one site, and the synapse uses one transmitter, acetylcholine, with one main receptor. A neuron in the brain differs on every count. A cortical pyramidal neuron receives thousands of synapses from many neurons. Excitatory synapses that use glutamate and inhibitory synapses that use GABA sit on the same dendrite, each with its own receptors. So a single signal that says “cluster receptors here” cannot be enough. Each contact has to specify what kind of synapse it will be.

A second observation from his lectures: neurons manufacture many synaptic building blocks before contact. In 1983 two groups showed that growth cones of cultured frog and chick neurons release acetylcholine before reaching a muscle, detected with a patch of receptor-rich membrane held near the growth cone. Presynaptic proteins travel down the axon in preassembled transport packets; in cultured hippocampal neurons, Stephen Smith’s laboratory watched packets carrying active-zone proteins arrive at new contacts. Dendrites carry small clusters of receptors and scaffold before any axon arrives.

His phrase for this: “Lonesome structures indicate that growth cones and dendrites are poised on the brink of differentiating into synaptic specialization.” And the rule that follows: “Presynaptic and postsynaptic specializations can form in isolation, but stable synapses require both.” Think of two halves of a bridge built out from opposite banks. Each half can stand for a while. Only joined do they carry traffic and last.

He also asks, “The decision to form a synapse: calcium signal?” In cultured spinal motor neurons, contact with a suitable partner cell raised free calcium in the growth cone, and a rise in cAMP could mimic part of the response. Whatever the first intracellular signal is, it has to be triggered by something on the partner’s surface or in the space between.

Signals that turn a growth cone into a bouton

Adhesion comes first. Cadherins are membrane proteins with five extracellular repeats that bind the same cadherin on the facing cell (homophilic binding) only when calcium is present. Nectins, held inside the cell by afadin, and SynCAM, a member of the immunoglobulin superfamily, also span the cleft. SynCAM made by a non-neuronal cell induced presynaptic terminals in contacting axons in a 2002 study. The question is which adhesion proteins only hold membranes together and which also instruct.

His example is the mossy fiber synapse of the cerebellum. Neurons in the pons, part of the brainstem, send axons called mossy fibers to cerebellar granule cells. The pontine growth cones carry neurexin, Frizzled and FGFR2. The granule cells make neuroligin, Wnt7a and FGF22. That gives three pairs:

1. Neurexin and neuroligin. Neurexins are axon-side membrane proteins; neuroligins are their dendrite-side partners. Scheiffele’s experiment above showed that neuroligin alone, on a cell that is not a neuron, makes pontine axons cluster vesicle proteins that recycle. Soluble beta-neurexin, which caps neuroligin, blocked the effect. The signal also runs the other way: Ann Marie Craig’s laboratory reported in 2004 that neurexin on a non-neuronal cell clusters postsynaptic proteins in dendrites that touch it.
2. Wnt7a and Frizzled. In 2000 Anna Hall, Frances Lucas and Patricia Salinas reported that Wnt7a from granule cells makes mossy fiber ends spread and branch and gather the vesicle protein synapsin I. In mice lacking Wnt7a, maturation of the mossy fiber terminals was delayed. Wnt7a acts through Frizzled receptors (see that lesson).
3. FGF22 and FGFR2. In 2004 Hisashi Umemori, Joshua Sanes and colleagues reported that FGF22, made by granule cells, promotes vesicle clustering in mossy fiber axons through FGFR2, a receptor tyrosine kinase (see that lesson), and that in mice, neutralizing FGF22 and its relatives FGF7 and FGF10, or inactivating FGFR2, reduced presynaptic differentiation of mossy fibers.

Design problem

You make a kidney cell line produce neuroligin and grow it under cultured pontine or hippocampal axons. What do you measure to show that an axon has built a real presynaptic terminal at the contact? Which two controls rule out mere adhesion?

Now the trap. “Sufficient in a dish” is not “required in the animal.” In 2003 Markus Missler, Thomas Südhof and colleagues reported mice lacking all three alpha-neurexins. They died soon after birth, and their synapses were present at close to normal density but released transmitter poorly. In 2006 mice lacking neuroligins 1, 2 and 3 were reported to have synapses in normal number and structure in the brainstem, with impaired transmission, more so at inhibitory synapses, and they died from failed breathing. So neurexins and neuroligins are required for normal synapse function and maturation, not for synapse formation. Many organizers overlap, and a dish holding one of them on a bare cell tests what that one can do, not what the brain depends on.

The receiving side: a scaffold that holds receptors

Under an excitatory terminal sits the postsynaptic density, the thickened membrane seen in electron micrographs. One of its most abundant proteins is PSD-95, a scaffold of the MAGUK family with three PDZ domains, short modules that grip the last few amino acids of a partner protein (see that lesson). The first two PDZ domains bind the tail of NMDA receptor GluN2 subunits. The third binds the tail of neuroligin. That gives a chain across the cleft: neurexin in the axon, neuroligin in the dendrite, PSD-95 beneath it, glutamate receptors held by PSD-95. A scaffold works like a pegboard: whatever has the right hook hangs in the right place (see that lesson).

AMPA receptors, the glutamate-gated channels that carry most fast excitatory current (see that lesson), do not bind PSD-95 directly. The clue came from a mouse. Stargazer is a spontaneous mutant that tosses its head upward and has seizures. Its gene encodes stargazin, a small membrane protein. In stargazer cerebellar granule cells, mossy fiber synapses give NMDA receptor responses but no AMPA receptor responses. In 2000 the laboratories of David Bredt and Roger Nicoll reported that stargazin binds AMPA receptors and brings them to the cell surface, and that its tail, which binds PSD-95’s PDZ domains, holds them at synapses. Stargazin became the first of the transmembrane AMPA receptor regulatory proteins, or TARPs. Other TARPs serve other neurons, which is one reason the stargazer defect shows most clearly in cerebellar granule cells.

Inhibitory synapses use a different scaffold. Gephyrin forms a lattice under the terminal that holds glycine receptors and many GABA-A receptors.

The signal for the receiving side can come from far away. In mouse sympathetic ganglia, Sharma, Deppmann, Harrington and colleagues reported in 2010 that nerve growth factor from the distant target tissue is needed for postsynaptic assembly in the ganglion neurons, which in turn is a prerequisite for presynaptic assembly, although preganglionic axons still grow into the ganglion without it. Signaling endosomes carrying its receptor TrkA travel back from the axon into the dendrites.

Astrocytes add the missing signal

Ben Barres’s laboratory could purify rat retinal ganglion cells by immunopanning, leaving cultures with almost no glia. In 1997 Frank Pfrieger and Barres reported that these neurons survived, grew and fired action potentials, yet barely spoke to one another. A layer of glia across the dish, not touching the neurons, or medium the glia had conditioned, changed that. With action potentials blocked by tetrodotoxin, the frequency of miniature synaptic currents rose roughly twelvefold, and failures of evoked transmission fell from about one stimulus in five to about one in fifty. Neuron survival did not change, so glia were not simply keeping more neurons alive.

In 2001 Erik Ullian and colleagues in the same laboratory reported that astrocytes raised the number of synapses about sevenfold and were needed to keep them. In 2005 Karen Christopherson and colleagues identified thrombospondins 1 and 2 as astrocyte-secreted proteins that induce synapses. Those synapses looked normal and released transmitter but were postsynaptically silent, with no AMPA receptor responses, and mice lacking both thrombospondins had fewer excitatory synapses. Later work identified other astrocyte factors, including glypicans 4 and 6, which recruit AMPA receptors and make such synapses respond, and cholesterol carried in lipoproteins. Structure and function come from different signals.

Keep the scope. Neurons form some synapses without astrocytes, and the size of the glial effect depends on the culture system. Astrocytes add to synapse formation; they are not its sole cause.

Maturation: GABA changes sign

A new synapse keeps changing after it forms. At the neuromuscular junction the acetylcholine receptor swaps its fetal gamma subunit for the adult epsilon subunit. In the brain, the best-known change is in GABA. In immature rodent hippocampal and cortical neurons, chloride inside the cell is high, because the transporter NKCC1 brings chloride in and the exporter KCC2 is scarce. Opening GABA-A channels then lets chloride flow out, and GABA depolarizes the cell. As KCC2 rises over the first postnatal weeks, internal chloride falls and GABA hyperpolarizes. The Nernst equation for chloride at 37 °C gives the switch:

E(Cl) = 61 mV × log10([Cl] inside / [Cl] outside)

With 30 mM chloride inside and 130 mM outside, E(Cl) is about −39 mV, above a resting potential near −60 mV, so opening the channel depolarizes. With 7 mM inside, E(Cl) is about −77 mV, and the same channel now inhibits.