Neural Induction: The Default Model

Cut a small square from the top of a frog blastula, culture it in salt water, and it becomes skin. Pull the same square apart into single cells, keep them apart for a few hours, then let them clump again, and they become neural tissue.

Both results come from the frog Xenopus laevis, and together they turned an old question around. A blastula is the early embryo, a hollow ball of cells. The square is the animal cap, the roof of the ball, and it is made of ectoderm: the outer germ layer, which gives rise to both the skin and the nervous system. Since the 1920s, when Hilde Mangold and Hans Spemann grafted the organizer (the tissue at the dorsal lip of the blastopore) and got a second body axis, the question had been what positive signal the organizer sends to make ectoderm neural. The square suggested another answer. Perhaps the signal in the ectoderm is the one that makes skin, and the organizer’s job is to stop it.

The course states it in one line: “Neural induction involves inhibition of BMP signals.” This lesson follows how that idea was built in the frog, which molecules carry it, and where it falls short.

Neural induction involves inhibition of BMP signals

Think of a room of singers who will all sing unless someone keeps saying “hush”. Put them in one small room and the hush reaches everyone. Spread them across a field and the hush fades, and they start to sing.

Ectoderm cells behave like the singers. In 1989 Grunz and Tacke reported that frog animal-cap cells separated from one another and kept apart for hours form neural tissue when they would otherwise have made epidermis. The cells are usually separated in medium that lacks the calcium and magnesium they need to stick together. Separation spreads the cells through a large volume of fluid. On the classic reading, whatever they normally pass to one another is diluted away; later work found that separated cells also switch on MAPK signaling, so dilution may not be the whole story.

In 1994 Ali Hemmati-Brivanlou and Douglas Melton reached the same place from the receptor side. They injected eggs with RNA for an activin receptor lacking its kinase domain. A receptor like this binds ligand and pairs with normal partners but cannot pass the signal on, so it jams the pathway; it is called a dominant negative. Animal caps from injected embryos became neural, with no mesoderm detected that could have induced them. The receptor had been built against activin, and the result was first read that way. It also blocks related ligands, and the reading shifted as a better candidate emerged.

That candidate was BMP4, a bone morphogenetic protein. BMPs belong to the TGF-beta family: a BMP dimer brings together type II and type I receptor kinases, and the type I receptor phosphorylates Smad1, 5 or 8, which travel to the nucleus with Smad4 (see that lesson). In 1995 Paul Wilson and Hemmati-Brivanlou added BMP4 to separated cap cells. The cells became epidermis, and their neural genes stayed off.

Put the three results side by side:

intact cap → epidermis; separated cells → neural; separated cells + BMP4 → epidermis

The conclusion, in this frog tissue, is that neural is the default fate, and it is actively suppressed by signaling between cells. Hemmati-Brivanlou and Melton set out this default model in a 1997 review. The lecture’s closing line puts the general lesson: “Neural Induction: A cell fate is determined in part by the signals to which it is exposed.”

Three antagonists catch BMP outside the cell

If BMP makes skin, the organizer should make something that blocks BMP. It makes at least three such proteins.

Noggin. In 1992 William Smith and Richard Harland found noggin by expression cloning, injecting RNA from pools of cloned genes into ultraviolet-treated embryos that lack dorsal structures and narrowing the pools to the one gene that rescued them. It encodes a secreted protein, and it is expressed in the organizer. In 1993 Teresa Lamb, Harland and colleagues showed that Noggin protein neuralizes animal caps without inducing dorsal mesoderm, so it acts on the ectoderm directly. In 1996 Lyle Zimmerman, José De Jesús-Escobar and Harland showed that Noggin binds BMP4 and keeps it off its receptors.

Chordin. In 1994 Yoshiki Sasai and colleagues in Eddy De Robertis’s laboratory cloned chordin as a gene switched on by organizer transcription factors. In 1996 Stefano Piccolo, Sasai, Bin Lu and De Robertis showed that Chordin binds BMP4 tightly and blocks its receptor binding. The Noggin and Chordin papers appeared side by side.

Follistatin. Known as a protein that binds activin, follistatin is made in the organizer, and in 1994 Hemmati-Brivanlou, O. G. Kelly and Melton showed that it neuralizes caps. It was later found to bind some BMPs as well.

Where do these proteins come from in the embryo? During gastrulation, the cell movements that form the germ layers, the organizer and the dorsal mesoderm turn inward and spread beneath the ectoderm. This is involution. The involuting mesoderm secretes Chordin, Noggin and Follistatin, and the ectoderm directly above it, cut off from BMP, becomes the neural plate. Chordin and Noggin are also made earlier, at the blastula stage, in dorsal animal cells, so neural induction starts before gastrulation. None of the three needs to enter a cell. Each acts on BMP in the fluid between cells, though follistatin does it differently: it can form a complex with BMP and its receptor instead of blocking receptor binding.

Chordin has a further twist, which the course calls the “Shuttling Mechanism”. Metalloproteases of the Tolloid family (Xolloid in the frog) cut Chordin and release the BMP it holds. In the model, a Chordin-BMP complex can travel farther than free BMP, so BMP is carried away from the organizer and set free where Chordin is cut, piling up on the ventral side. The antagonist both lowers BMP signaling dorsally and helps shape a gradient that peaks ventrally.

This Is Familiar: the same contest in a fly, upside down

The fruit fly has the same pair. Its gene decapentaplegic (dpp) encodes a BMP2/4 relative, and short gastrulation (sog) encodes a relative of Chordin. In 1995 Holley and colleagues, a collaboration of fly and frog laboratories including De Robertis’s, reported that the two antagonists act alike when swapped between species: frog Chordin works in the fly and fly Sog works in the frog. Tolloid itself was first named as a fly gene, and it cuts Sog.

The map is inverted. In the fly, Dpp is made on the dorsal side and Sog on the ventral side, and the nerve cord forms ventrally, where Sog keeps Dpp signaling low. In a vertebrate, BMP4 signaling is high on the ventral side, Chordin comes from the dorsal organizer, and the neural tube forms dorsally. Turn a fly over, and its BMP and antagonist lie where a frog’s do. In 1996 De Robertis and Sasai argued that the common ancestor of flies and vertebrates already patterned its body this way, and that one lineage inverted the plan. Étienne Geoffroy Saint-Hilaire had proposed an inversion of the body plan between arthropods and vertebrates in the 1820s on anatomy alone.

Where the default model strains

One inducer or two instructions? Noggin-treated caps made markers of cement gland and forebrain, but not of hindbrain or spinal cord. Neural tissue made by blocking BMP is anterior. That fits an older result. In 1952 Pieter Nieuwkoop and coauthors at the Hubrecht Laboratory in Utrecht reported experiments in which a fold of responsive amphibian ectoderm, planted with one end in a host’s neural plate, could form posterior structures near the attachment and forebrain farther along it. Nieuwkoop proposed two steps: activation, which makes ectoderm neural and anterior, and transformation, which shifts part of it toward hindbrain and spinal cord. In molecular terms the BMP antagonists look like activators. Wnt signals (see that lesson) and FGFs are leading candidates for transformation. Mouse stem-cell work has since suggested that much of the spinal cord comes from progenitors that take on a posterior identity before they become neural, so the order may differ from brain to cord.

In chick, blocking BMP is not enough. Claudio Stern’s laboratory reported in 1998 that misexpressing Chordin in chick regions outside the future neural plate does not induce early neural markers, although Chordin did stabilise early neural markers in cells that had already received neural-inducing signals. In 2000 Andrea Streit and colleagues reported that FGF signaling before gastrulation is needed to begin neural induction in chick. In 2001 Sara Wilson and colleagues in Thomas Edlund’s laboratory reported that in chick epiblast the state of Wnt signaling decides whether cells respond to FGF by becoming neural or epidermal. In mouse, embryos lacking both chordin and noggin show severe defects in forebrain development. A fair summary is that BMP inhibition is needed for normal neural development in several vertebrates, and that outside the frog cap it is not sufficient.

None of this shows the frog results were wrong. All the animal-cap work is in Xenopus laevis, and it should not be stretched to mammals.

Design problem

You have Xenopus animal caps and five conditions. Predict the fate of each: (1) an intact cap; (2) an intact cap given Noggin protein; (3) cells separated for three hours, then reaggregated; (4) separated cells given BMP4; (5) an intact cap from an embryo injected with RNA for a dominant-negative BMP receptor. Then say which single result would have falsified the default model.