Developmental Neuroscience · Unit 3: Coordinates: patterning the tube

Ventral to Dorsal: Sonic Hedgehog and BMP in the Neural Tube

How does a dividing cell in the neural tube know how far it sits from the bottom, and so whether its daughters become motor neurons or relay neurons for touch and pain? This lesson follows Sonic hedgehog from the notochord and floor plate below, and BMPs from the epidermis and roof plate above, and shows how progenitors read the level and the duration of the ventral signal and sharpen a smooth slope into sharp domains. Afterwards you can predict the result of grafting, removing or blocking each source, and of breaking one of the transcription factors that read it.

Watch first A Slope Becomes Stripes · 2:44

Graft a second notochord against the side of a chick embryo’s neural tube, and the neural cells beside it start to behave like floor plate. The tube gains a second bottom.

The neural tube is the hollow rod that rolls up from the neural plate, a sheet of ectoderm (the embryo’s outer cell layer), and becomes the brain and spinal cord. Its bottom strip is the floor plate, a narrow line of specialized cells. Its top strip is the roof plate. Under the tube runs the notochord, a rod of mesoderm. In the wall between floor and roof, dividing cells called progenitors will make motor neurons near the bottom and relay neurons for touch and pain near the top.

So each progenitor needs to know how far it sits from the bottom. The answer starts with a sentence Chris uses for this part of the course: “Differentiation in the neural tube depends on factors from adjacent, non-neural, tissues.”

The idea: the tube learns its bottom and top from its neighbours

Notochord and floor plate lie against each other and share features. That likeness could mean a common history, or it could mean the notochord tells the neural cells what to become. A static embryo cannot tell the two apart. Moving tissue can.

An earlier group had seen an extra floor plate after implanting a notochord in chick embryos in 1985. In 1990 Marysia Placzek, Marc Tessier-Lavigne, Toshiya Yamada, Thomas Jessell and Jane Dodd ran the test in both directions in chick. A notochord grafted beside the lateral wall of the tube induced floor-plate properties in the neural cells it touched. Removing the normal notochord early prevented floor-plate properties at the midline. The first result shows sufficiency, the second necessity.

The floor plate then passes the message on. In a 1993 culture assay, Placzek, Jessell and Dodd found that floor plate induced floor-plate markers in neural plate explants as well as notochord did, and that the signals acted by contact. They called this homeogenetic induction: an induced tissue recruits its neighbours to the same identity.

The top of the tube has its own neighbours. The epidermis, the non-neural ectoderm on either side of the closing neural plate, and later the roof plate send dorsalizing signals. Chris’s summary of the axis is short. BMPs (bone morphogenetic proteins) act in neural induction and in dorsal identity, and dorsal BMP and ventral Sonic hedgehog work against each other.

Hedgehog arrives: “Shh protein is sufficient to turn an explant ventral”

The signal got a name in the December 1993 issue of Cell. Three papers described vertebrate relatives of hedgehog, a fly gene for a secreted signal: Yann Echelard, Andrew McMahon and colleagues in mouse; Stefan Krauss, Jean-Paul Concordet and Philip Ingham in zebrafish; and Robert Riddle, Randy Johnson, Ed Laufer and Cliff Tabin in chick. The gene they called Sonic hedgehog (Shh) was expressed in notochord and floor plate, and in the mouse paper, forcing it on where it is normally absent in the central nervous system switched on floor-plate genes.

Jessell’s laboratory then put the protein back into the assay. An explant is a small piece of tissue cut out and grown in a dish. In 1994 Henk Roelink and colleagues showed that cultured cells engineered to make a vertebrate Hedgehog induced floor-plate and motor-neuron markers in neural plate explants. In 1995 they used the purified amino-terminal fragment, the part of the protein that signals, and found that higher concentrations induced floor plate and lower ones induced motor neurons. One protein, at different doses, gave different ventral fates.

Necessity came from a knockout, a mouse with a gene deliberately disrupted. In 1996 Chin Chiang, Philip Beachy and colleagues reported that mice lacking Shh have no floor plate and lack most ventral cell types of the neural tube.

Inside the cell, Shh binds its receptor Patched, which releases Smoothened, and the Gli transcription factors stop being cut into repressors at the primary cilium (see that lesson). What a progenitor reads, in this model, is the balance of Gli activator against Gli repressor. Two links to people make the pathway’s weight clear. Heterozygous mutations in human SHH can cause holoprosencephaly, in which the forebrain fails to divide into two halves. And cyclopamine, an alkaloid from the plant Veratrum that causes cyclopic lambs when pregnant ewes graze on it, blocks Smoothened.

Reading a gradient by level and by time

A morphogen is a signal that spreads from a source and gives different instructions at different concentrations. Shh from the floor plate falls off with distance, so progenitors at different heights receive different amounts. Their answer is written in homeodomain transcription factors.

In 2000 James Briscoe, Alessandra Pierani, Jessell and Johan Ericson grouped these factors in chick into two classes. Class I proteins, including Pax7, Dbx1, Dbx2, Irx3 and Pax6, are repressed by Shh, so each is expressed above a border and lost below it. Class II proteins, Nkx6.1 and Nkx2.2, need Shh, so each is expressed below a border. Pax7 is switched off by low levels and so marks the dorsal half; Nkx2.2 needs high levels and sits next to the floor plate. Their staggered borders define five ventral progenitor domains. From dorsal to ventral these are p0, p1, p2, pMN and p3, which make V0, V1 and V2 interneurons, motor neurons and V3 interneurons. Olig2, a bHLH repressor rather than a homeodomain protein, marks pMN and acts with these factors (Novitch et al. 2001). The markers Chris shows, Pax7, Olig2 and Nkx2.2, read the axis from top to bottom.

Concentration turned out to be only half of the reading. In 2007 Eric Dessaud, Briscoe and colleagues reported that chick neural cells adapt: Shh switches on its own receptor Patched1, which makes cells less sensitive, so signaling inside the cell fades. Higher concentrations keep signaling going longer. A 2010 study from the same laboratory held the dose fixed in chick explants and varied exposure time, and longer exposure gave progressively more ventral identities. Duration carries information in many pathways (see that lesson). Here, in this model, a cell’s identity tracks something like the total signaling it has received over time.

Turning a slope into stripes: cross-repression

A smooth gradient should give a blur, yet the domains have sharp borders. The reason is that factors on either side of a border repress each other. Briscoe and colleagues showed in chick by electroporation that Pax6 and Nkx2.2 repress each other at the border between pMN and p3, and that Dbx2 and Nkx6.1 do the same at the border between p1 and p2. A later study (Chen et al., Neuron 2011) described a cross-repressive loop between Irx3 and Olig2 at the dorsal edge of pMN. Forcing Nkx2.2 into pMN suppresses motor neurons and gives V3 neurons.

Two factors that repress each other form a switch with two stable states, the bistable circuit from that lesson. A cell near a border cannot hold both. It falls to one side, and stays there.

The network also shapes where the border falls. In 2012 Nikolaos Balaskas, Vanessa Ribes, Briscoe and colleagues studied the mouse neural tube with a fluorescent reporter of Gli activity. In mice lacking both Pax6 and Olig2, Nkx2.2 spread dorsally although the reporter range did not. When embryos in culture received a short pulse of a drug that activates Smoothened, the borders stayed close to their normal positions; continuous treatment shifted them. The network buffers a passing change and remembers enough of its past to keep a border. Whether the gradient is read directly or mainly through such feedback is still argued.

The dorsal half: BMPs from the roof plate

The top of the tube runs a parallel system with different signals. In 1995 Karel Liem, Gabi Tremml, Henk Roelink and Jessell found that BMP4 and BMP7 could mimic the dorsalizing effect of epidermis on chick neural plate explants. In 1997 Liem, Tremml and Jessell proposed that BMPs from the epidermis induce the roof plate, which becomes a second source of related signals. BMPs are members of the TGF-beta family and act through receptor kinases and Smad proteins (see that lesson). The roof plate also makes Wnts, including Wnt1 and Wnt3a.

The dorsal tube makes six early classes of interneurons, named dI1 to dI6 from the top down. In 2000 Kevin Lee, Peter Dietrich and Jessell removed the roof plate genetically in mice, and the three dorsal-most classes later failed to form. The two systems also oppose each other: BMP signaling alters the response to Shh, and BMP inhibitors such as follistatin, made by the notochord and somites, help shift cells toward ventral fates (Liem, Jessell and Briscoe, 2000).

Design problem

You have intermediate neural plate explants from chick embryos, which on their own express Pax7. You also have purified Shh signaling fragment, cyclopamine and antibodies to Pax7, Olig2 and Nkx2.2. Design an experiment that tells whether progenitors read the Shh concentration, the duration of exposure, or both.

One way to do it

Run two series and score every explant at the same endpoint, say 48 hours, so age is the same for all.

In the dose series, give each explant a different concentration for the whole time. If concentration matters, rising dose should first switch off Pax7, then switch on Olig2, then at the highest doses switch on Nkx2.2 and lose Olig2.

In the duration series, hold one intermediate dose fixed and add cyclopamine at different times to end signaling. If duration matters, short exposures should leave cells Pax7-negative but without Olig2, longer ones should give Olig2, and with a higher fixed dose the longest exposures should also give Nkx2.2.

The expected answer from chick explant studies is that both matter. At a fixed time, higher doses push cells ventrally; at a fixed dose, longer exposure does too. Add one more test: measure a Gli reporter over time at each dose. Signaling should fade as cells adapt, faster at low dose, which is how one variable becomes two.

How we measure it

Tissue grafts and ablations in chick embryos

A piece of tissue, such as a notochord, is moved to a new position in a living embryo, or removed. Gain at the new site tests sufficiency; loss at the normal site tests necessity. Chick embryos suit this because they develop in an egg that can be opened and resealed.

Neural plate explants with purified signal

A small piece of neural plate is cut out and grown in a gel with a known concentration of a protein such as the Sonic hedgehog signaling fragment, for a set time. Antibodies to transcription factors then show which identity the cells took, so dose and duration can be varied one at a time.

In ovo electroporation

DNA is injected into the chick neural tube and pushed into cells on one side with brief electric pulses. The other side serves as a control in the same embryo, which is how cross-repression between pairs of transcription factors was tested.

Mouse knockouts and reporter lines

A knockout mouse carries a gene deliberately disrupted, so its embryos show what the gene is needed for. A reporter line carries a fluorescent protein driven by the target sites of a pathway, here Gli-binding sites, so signaling can be watched in the embryo.

Finish line

Test yourself

8 questions

Practice and MCAT-style, with explained answers.

Next · Lesson 11

Transcription Factor Codes: Building a Motor Neuron

Two transcription factors, two cell types: swap one partner and the same protein builds an interneuron instead.

Continue →

← Lesson 9 · Local Organizers: The Midbrain-Hindbrain Boundary

Sonic hedgehog signaling domain — the morphogen itself
Sonic hedgehog signaling domain — the morphogen itself ● LIVE · RCSB PDB 1VHH
open on RCSB PDB ↗
Loading live structure 1VHH from RCSB PDB…

This is the fragment that remains after the protein cuts itself in two. Look for the zinc ion held in a cleft on one face, a site that resembles the active site of a zinc peptidase.

Drag to rotate · scroll or pinch to zoom · right-drag to pan
GLI zinc fingers on DNA — the reader inside the nucleus
GLI zinc fingers on DNA — the reader inside the nucleus ● LIVE · RCSB PDB 2GLI
open on RCSB PDB ↗
Loading live structure 2GLI from RCSB PDB…

Five small zinc-held modules wind along the DNA. Not all of them make the same contacts, which you can see by following which fingers sit deep in the major groove.

Drag to rotate · scroll or pinch to zoom · right-drag to pan
BMP-7 — a dorsal signal from the roof plate
BMP-7 — a dorsal signal from the roof plate ● LIVE · RCSB PDB 1BMP
open on RCSB PDB ↗
Loading live structure 1BMP from RCSB PDB…

Two chains join through a disulfide bond, each folded around a knot of disulfides typical of the TGF-beta family. The dimer's two curved arms are where receptors bind.

Drag to rotate · scroll or pinch to zoom · right-drag to pan
Cyclopamine — the plant alkaloid that blocks Smoothened
Cyclopamine — the plant alkaloid that blocks Smoothened ● LIVE · PubChem CID 442972
open on PubChem ↗
Loading live structure 442972 from PubChem…

A rigid steroid-like ring system with a nitrogen-containing end. It binds Smoothened and blocks Hedgehog signaling, which is why it is used to switch the pathway off in explants.

Drag to rotate · scroll or pinch to zoom · right-drag to pan
Marysia Placzek

Marysia Placzek

With Marc Tessier-Lavigne, Toshiya Yamada, Thomas Jessell and Jane Dodd, showed in chick that a grafted notochord induces floor plate in the side of the neural tube and that removing the notochord early prevents the normal floor plate, then showed that induced floor plate can itself induce floor plate.

Notochord and floor plate lie against each other and share features, so their likeness could mean a shared history or an instruction passed from one to the other. In 1990, working with Marc Tessier-Lavigne, Toshiya Yamada, Thomas Jessell and Jane Dodd, Placzek tested both directions in chick embryos. A notochord grafted beside the lateral wall of the neural tube gave the nearby neural cells floor-plate properties, and removing the normal notochord early prevented floor-plate properties at the midline. An earlier group had already seen an extra floor plate after a notochord implant; the 1990 paper added the test of necessity. In 1993, with Jessell and Dodd, she placed neural plate explants against notochord or floor plate in culture. The signals acted by contact, and floor plate worked as an inducer as well as notochord. They called this homeogenetic induction: an induced tissue recruits its neighbours to the same identity. That gave a cellular account of how a narrow contact could build a long midline strip, and it set up the search for the signal.

✓ verbatim Placzek, Tessier-Lavigne, Yamada, Jessell & Dodd, Science 250:985 (1990); Placzek, Jessell & Dodd, Development 117:205 (1993).
Andrew McMahon

Andrew McMahon

With Yann Echelard and colleagues, cloned mouse Sonic hedgehog, found it in notochord, floor plate and the limb's polarizing region, and showed that expressing it in the wrong part of the mouse central nervous system switches on floor-plate genes.

By 1993 the fly gene hedgehog was known to encode a secreted signal that patterns segments, and several laboratories searched vertebrate genomes for relatives. Andrew McMahon's laboratory, with Yann Echelard as first author, found three mouse genes and named one Sonic hedgehog. Its RNA lay in exactly the tissues the graft experiments had pointed to: the notochord, the floor plate and the zone of polarizing activity, the small patch at the back of the limb bud that sets the order of the digits. When the team expressed Sonic hedgehog in transgenic mice in parts of the central nervous system where it is normally absent, genes typical of floor plate turned on there. The paper appeared in the same issue of Cell as two others: Stefan Krauss, Jean-Paul Concordet and Philip Ingham with the zebrafish gene, and Robert Riddle, Randy Johnson, Ed Laufer and Cliff Tabin with the chick gene, which reproduced the limb's polarizing activity. Three species and one expression pattern made Sonic hedgehog the leading candidate for the notochord's signal, ready to be tested as a protein.

✓ verbatim Echelard et al., Cell 75:1417 (1993); Krauss, Concordet & Ingham, Cell 75:1431 (1993); Riddle, Johnson, Laufer & Tabin, Cell 75:1401 (1993).
Thomas Jessell

Thomas Jessell

His laboratory showed that the amino-terminal fragment of Sonic hedgehog induces floor plate at high concentration and motor neurons at lower concentration, defined the class I and class II homeodomain code of ventral progenitors, and identified BMPs from the epidermis and roof plate as dorsal signals.

The laboratory of Thomas Jessell at Columbia University took the floor plate from an anatomical seam to a coordinate system. After the notochord grafts, the group, with Henk Roelink, showed in 1994 that cells engineered to make a vertebrate Hedgehog induce floor-plate and motor-neuron markers in neural plate. In 1995 Roelink and colleagues used the purified signaling fragment and found that the higher concentrations they tested induced floor plate and lower ones induced motor neurons. With Johan Ericson and James Briscoe the laboratory then mapped the transcription factors that progenitors switch on or off in response, and in 2000 Briscoe, Alessandra Pierani, Jessell and Ericson proposed a code of class I and class II proteins that defines five ventral progenitor domains. In the same years Karel Liem, Gabi Tremml and Jessell showed that BMPs can mimic the dorsalizing effect of epidermis on neural plate and that the roof plate becomes a second source of related signals. The work gave the spinal cord a map that later studies of motor neuron identity and circuits could build on.

✓ verbatim Roelink et al., Cell 81:445 (1995); Briscoe, Pierani, Jessell & Ericson, Cell 101:435 (2000); Liem, Tremml & Jessell, Cell 91:127 (1997).
James Briscoe

James Briscoe

Showed with Eric Dessaud and colleagues that chick neural cells adapt to Sonic hedgehog, so that both its concentration and the duration of signaling set identity, and with Nikolaos Balaskas, Vanessa Ribes and colleagues that a small gene network in the mouse neural tube turns the gradient into stable domains.

Briscoe worked on the ventral code in Jessell's laboratory, as an author of the 1997 and 2000 papers that tied graded Sonic hedgehog to Pax6, Nkx2.2 and their relatives. A threshold diagram of that code invited a question: does a cell simply measure how much ligand it sees? In 2007, with Eric Dessaud and colleagues, his own laboratory in London reported that chick neural cells convert different concentrations into periods of intracellular signaling of different length, because the pathway desensitizes them over time, partly through rising Patched1. Higher concentrations kept signaling going longer. A 2010 study, with Dessaud and Noriaki Sasai, separated level from exposure time in chick explants and found that longer exposure gave progressively more ventral identities. In 2012, with Nikolaos Balaskas, Vanessa Ribes and colleagues, the laboratory turned to mouse embryos and a Gli reporter, and showed that the interactions among Pax6, Olig2 and Nkx2.2 help decide where boundaries sit and keep them in place after a brief change in signal. The work made the gradient a problem in dynamics.

✓ verbatim Dessaud et al., Nature 450:717 (2007); Balaskas et al., Cell 148:273 (2012).
Test yourself
Question 1 of 8 Score 0 / 0

Which tissue first supplies Sonic hedgehog to the ventral neural tube?
Sources & Integrity

Every structure, slide, simulation, and clip on this page is listed below with the source it came from. None of the 3D structures is drawn by hand or by AI: each one loads by accession ID from RCSB PDB or PubChem. This panel is generated from the lesson's own data file, so what it lists is what the page loads.

Molecular structures live data

  • Sonic hedgehog signaling domain — the morphogen itself — RCSB PDB 1VHH · coordinates CC0 · Mouse Sonic hedgehog amino-terminal signaling domain, X-ray crystal structure (Hall et al., Nature, 1995)
  • GLI zinc fingers on DNA — the reader inside the nucleus — RCSB PDB 2GLI · coordinates CC0 · Five zinc fingers of human GLI1 bound to DNA, X-ray crystal structure (Pavletich & Pabo, Science, 1993)
  • BMP-7 — a dorsal signal from the roof plate — RCSB PDB 1BMP · coordinates CC0 · Human osteogenic protein-1 (BMP-7) dimer, X-ray crystal structure (Griffith et al., PNAS, 1996)
  • Cyclopamine — the plant alkaloid that blocks Smoothened — PubChem CID 442972 · Cyclopamine, a steroidal alkaloid from Veratrum californicum (PubChem compound record)

Scientist profiles & quotes

  • Marysia Placzek — Placzek, Tessier-Lavigne, Yamada, Jessell & Dodd, Science 250:985 (1990); Placzek, Jessell & Dodd, Development 117:205 (1993).
  • Andrew McMahon — Echelard et al., Cell 75:1417 (1993); Krauss, Concordet & Ingham, Cell 75:1431 (1993); Riddle, Johnson, Laufer & Tabin, Cell 75:1401 (1993).
  • Thomas Jessell — Roelink et al., Cell 81:445 (1995); Briscoe, Pierani, Jessell & Ericson, Cell 101:435 (2000); Liem, Tremml & Jessell, Cell 91:127 (1997).
  • James Briscoe — Dessaud et al., Nature 450:717 (2007); Balaskas et al., Cell 148:273 (2012).

Quotes are verbatim from the cited source. Profile text is original to this site and was checked against the sources cited.

Animated explainer

A Slope Becomes Stripes · narration: an AI narrator voice (Gemini, Charon) reading narration written from this lesson's text · animation: AI-generated in code (Manim) by Claude, a schematic with no drawn molecular structures, plots computed from their equations · captions from the narration script.

Images & figures

This lesson carries no raster figures. Every structure is rendered live from RCSB PDB or PubChem, and every plot is drawn from its equation in the browser.

integrity Machines check the mechanics. The build confirms that every local file exists, that every question's answer key points at a real choice, and that any AI-generated scientific figure names the person who checked it; a weekly job re-resolves every accession ID against its database. Questions are original; MCAT-style items are written for this site and are not real AAMC questions. Banned sources (the Tymoczko textbook, publisher decks, copyrighted figures) are not used.

Like I'm five

Picture a tall tube with a lamp at the bottom and a different lamp at the top. Each cell in the wall looks at how much light it gets from each lamp, and for how long. Cells near the bottom lamp become one kind of nerve cell, cells near the top become another, and cells in between become the kinds in the middle. Neighbours that pick different jobs push each other apart, so the stripes come out neat.
High school

The spinal cord starts as a hollow tube in the embryo. A rod of tissue under the tube, the notochord, releases a protein called Sonic hedgehog, and it turns the bottom strip of the tube into the floor plate, which then releases the same protein. The amount of Sonic hedgehog falls with distance from the bottom, so cells at different heights receive different amounts. Each cell switches certain genes on or off depending on how much signal it gets and for how long, and that sets which kind of neuron it will make, such as motor neurons near the bottom. Meanwhile, the skin above and the top strip of the tube release different signals, BMPs, that make the top half produce other kinds of neurons.
College

The notochord induces the floor plate, and both secrete Sonic hedgehog (Shh), which forms a ventral-to-dorsal gradient. The epidermis and later the roof plate secrete BMPs and Wnts, which promote dorsal identity, and the two systems oppose each other. Ventral progenitors read Shh through the balance of Gli activator and Gli repressor, set at the primary cilium. Their response is a set of homeodomain factors: class I proteins (Pax7, Irx3, Pax6, Dbx1, Dbx2) are repressed by Shh, and class II proteins (Nkx6.1, Nkx2.2) are induced. Pairs with adjacent borders repress each other, such as Pax6 and Nkx2.2, or Irx3 and Olig2, so the graded input resolves into the p0, p1, p2, pMN and p3 domains that make V0, V1 and V2 interneurons, motor neurons and V3 interneurons. Dorsally, BMP and Wnt signaling from the roof plate is needed for the dorsal-most interneuron classes among dI1 to dI6.
Graduate

In chick explants, Shh-N concentration thresholds in the low nanomolar range distinguish the induction of Nkx6.1, the repression of Irx3 and the induction of Nkx2.2. Concentration is not read as a static value. Shh induces Ptch1, which desensitizes cells, so in chick neural tissue a higher concentration produces a longer period of Gli activity, and at a fixed dose longer exposure yields more ventral identities; Dessaud and colleagues summarized this as positional identity tracking something like the time integral of signaling. In the neural tube, a network in which Pax6 represses Nkx2.2, Olig2 represses Pax6 and Nkx2.2 represses both Pax6 and Olig2 shows hysteresis: in chick explants less signaling maintains Nkx2.2 than is needed to induce it, and in cultured mouse embryos the pattern is little changed by a transient pulse of a Smoothened agonist.

An open question is how much of the final pattern reflects the shape of the extracellular gradient and how much the downstream network. Domain boundaries are more precise than the measured gradient, Gli reporters report signaling with a delay, and opposing BMP signaling and tissue growth change what each cell sees over time. Whether the gradient is read directly, or mainly through feedback and cross-repression, is still argued.
Three hedgehogs in one issue

The December 1993 issue of Cell carried three papers in a row on vertebrate relatives of the fly gene hedgehog. Cliff Tabin's group reported the chick gene and showed that cells making it could reproduce the limb's polarizing activity. Andrew McMahon's group reported the mouse genes and found Sonic hedgehog in notochord and floor plate. Stefan Krauss, Jean-Paul Concordet and Philip Ingham reported the zebrafish gene in the same midline tissues. The name Sonic, after the video-game character, stuck. Consecutive pages record a field that converged quickly, not who arrived first.

— Riddle et al., Cell 75:1401 (1993); Echelard et al., Cell 75:1417 (1993); Krauss, Concordet & Ingham, Cell 75:1431 (1993).
Induction by contact

Placzek, Jessell and Dodd asked whether notochord and floor plate need to touch neural tissue to induce floor plate. Their 1993 abstract reports that contact-mediated signals from both tissues act directly on neural plate cells, and that floor plate can pass the induction on to more lateral cells. The result separated a short-range contact signal from a freely diffusing one.

— Placzek, Jessell & Dodd, Development 117:205 (1993).

 ↑ Top