Transcription Factor Codes: Building a Motor Neuron

A progenitor in the ventral spinal cord does not contain a gene for “motor neuron”. It contains three or four transcription factors, none of which is specific to motor neurons on its own, and a motor neuron is what that particular set spells.

This is the clearest worked example in development of an idea you have met in outline: signals restrict which transcription factors a cell makes, and those factors then regulate target genes. Here you can follow it the whole way, from a stripe across a tube to a cell that holds a muscle fibre at the other end of a long axon. You can also run the logic backwards, which is how motor neurons are now made in a dish.

Two warnings before you start. Much of the evidence comes from chick embryos, where a plasmid can be driven into one side of the neural tube and the other side serves as the control, and much of the rest comes from mouse deletions. The two ask different questions, and they are easy to confuse. Through the lesson, note which animal and which manipulation a claim rests on.

From a stripe of progenitors to one cell type

Sonic hedgehog from the notochord and then the floor plate sets up a gradient across the neural tube, opposed by BMPs from above, and progenitors read their position in it; the Hedgehog lesson covers how that signal is received. The read-out is a set of stacked domains, each with its own transcription-factor profile. One of them, called pMN, makes motor neurons.

A pMN progenitor is still dividing, and it carries the bHLH repressor Olig2 along with Nkx6.1 and Pax6. Novitch, Chen and Jessell reported in 2001 that Olig2 marks these progenitors and, with Neurogenin2, couples the acquisition of motor-neuron character to becoming a neuron at all. Mizuguchi and colleagues, misexpressing the two factors in chick, reached a similar conclusion in the same year.

How much does the domain depend on Olig genes? In mice lacking both Olig1 and Olig2, motor neurons are largely eliminated and oligodendrocyte differentiation is abolished; lineage tracing in that study suggested the same progenitors instead make V2 interneurons and then astrocytes. Two groups reported the double mutant in 2002. Keep the scope in mind: this is the mouse spinal cord, and Olig2 is the one required for motor-neuron and oligodendrocyte specification in the cord, while Olig1 contributes to oligodendrocyte development and maturation, especially in the brain. The same pMN domain later turns to making oligodendrocytes, which the glia lesson takes up.

Around its final division a motor neuron turns on a new set: Isl1 or its relative Isl2, Lhx3, and Mnx1, the gene also called Hb9. These are the markers you would stain for to find a motor neuron in a section.

A code, not a master gene

Here is the experiment that makes the word “code” more than a metaphor.

Lhx3 is expressed both in motor neurons and in their neighbours, the V2 interneurons. In the chick neural tube, put Lhx3 into cells that have no Isl1 and you get V2 interneuron genes. Put Isl1 and Lhx3 into the same cell and you get motor-neuron genes. Thaler and colleagues in Pfaff’s laboratory explained why in 2002. Lhx3 binds the LIM cofactor NLI (also called Ldb1) directly, and that complex triggers V2 differentiation. In a cell that also makes Isl1, Isl1 competes for NLI and displaces Lhx3 onto a high-affinity site in the Isl1 C-terminal region. The complex that forms then promotes motor-neuron genes.

So the cell is not reading a gene that says motor neuron. It is reading which assembly wins a competition for a shared partner. Compare the two structures in this lesson: the same surface of Lhx3 holds NLI in one and a segment of Isl1 in the other.

The sufficiency and necessity logic falls out cleanly. Ectopic expression in chick tells you what a factor can impose: Mnx1 or its chick relative MNR2, electroporated into the cord, is sufficient to trigger motor-neuron differentiation and to repress the neighbouring V2 programme. Deletion in mouse tells you what cannot be built without it: without Isl1, motor neurons are not generated at all, while without Mnx1 they are generated on schedule and in normal numbers but transiently take on V2 features and end up with disordered columns and axons. Being needed to make a cell and being needed to keep it are different jobs.

Design problem

You can electroporate chick dorsal spinal cord, where Lhx3 is absent and Isl1 is confined to a small intermediate population of interneurons, and you can also look at an Isl1 knockout mouse. Predict the marker result for each of four conditions: (a) Lhx3 alone; (b) Isl1 plus Lhx3; (c) a single fused Isl1-Lhx3 protein; (d) motor neurons in the Isl1 knockout mouse.

Where along the cord: Hox genes, columns and pools

A motor neuron in your neck and one in your lower back were built by the same code, yet one drives a neck muscle and the other a muscle of the leg. The difference is rostrocaudal, and it is written in Hox proteins, the same family you met setting rhombomere identity in the hindbrain.

Dasen, Liu and Jessell reported in 2003 that sequential phases of Hox-c activity control columnar fate in motor neurons. Graded fibroblast growth factor signalling establishes Hox expression in progenitors, which is translated into a Hox pattern in motor neurons, where cell autonomous repressor and activator functions sort cells into columns. A 2005 paper from the same laboratory described the network in more detail: one set of Hox interactions fixes position along the cord, and another generates diversity among pools within a single segment, with the pattern of connections to target muscles as the read-out.

The columns are defined by where the cell body sits and where the axon goes. Motor neurons of the medial motor column innervate axial muscles of the back. Those of the lateral motor column, found at limb levels, innervate limb muscles. Those of the preganglionic column project to autonomic ganglia. Within a column, a pool is the set of neurons that innervates one muscle. Do not expect a tidy count of pools; the number depends on the species and on how finely you choose to divide.

FoxP1 is the switch on all of this. Dasen and colleagues reported in 2008 that FoxP1 acts as a Hox accessory factor, expressed in the Hox-sensitive columns and acting in a dose-dependent way, and Rousso and colleagues in Novitch’s laboratory showed in the same year that Foxp1 defines columnar identity and sets the pattern of LIM homeodomain proteins. Delete Foxp1 in mouse and the output of the motor-neuron Hox network is abolished, and in the limb motor axons appear to select trajectories and muscle targets at random. Motor neurons are still made; what fails is the assignment of which one each is.

Retinoic acid appears here too, and not only as a laboratory reagent. Sockanathan and Jessell reported in 1998 that a retinoid signal from one subset of early-born motor neurons helps specify a later-born subset, so some positional information comes from the neurons themselves. For how retinoic acid is received, see nuclear receptors.

Which half of the limb: the LIM code sets the first axon choice

The lateral motor column splits again, and the split is the clearest case of a transcription factor code handing off to axon guidance. Lateral LMC neurons express Lim1 (Lhx1); medial ones retain Isl1. Kania, Johnson and Jessell showed in mouse in 2000 that Lim1 in the lateral set ensures their axons take a dorsal trajectory in the limb, while Lmx1b in dorsal limb mesenchyme controls both trajectories from outside. Without the two proteins, motor axons appear to select dorsal and ventral paths at random. The downstream machinery is a set of Eph receptors and ephrins, which guidance cues and pioneers and topographic maps take up.

The order is worth holding onto. Position in the tube sets the progenitor domain. The domain sets the postmitotic code. Position along the cord sets the column. The column sets which receptors the axon carries. Only then does the axon meet the gradient that steers it.

Making motor neurons in a dish

Run the pathway forward and you have a protocol. Wichterle and colleagues reported in 2002 that mouse embryonic stem cells, given the signals that pattern the embryonic cord, became spinal progenitors and then motor neurons along a route that recapitulated the embryonic one. The cells populated an embryonic spinal cord after transplantation, extended axons and formed synapses with target muscles.

Two added molecules do most of the work, and each stands for one axis. Retinoic acid pushes the cells to a caudal, spinal character rather than a forebrain one. A Hedgehog agonist, either Shh protein or a small molecule such as purmorphamine or SAG, ventralises them toward pMN. Then you wait, and Mnx1-positive neurons appear.

Later work added control over subtype. Adams and colleagues in 2015 programmed limb-innervating LMC neurons from mouse and human embryonic stem cells by manipulating Foxp1, and those cells preferentially projected toward limb muscle. That is the Hox-and-FoxP1 layer, applied on purpose.

What the dish still does badly is the fine grain. You can get motor neurons, and you can bias them toward a column. Getting one named pool, the cells that innervate one muscle, is another matter, which is a fair measure of how much of the code is still unwritten.