Getting to the Muscle
The Spinal Cord, the Descending Pathways, and the Motor Cortex
The final common path
We ended the overview with a promise to begin where movement actually meets muscle. So let us begin there, with a fact so simple it is easy to walk past: whatever the brain decides to do, the only way it can move the body is by contracting muscles, and the only cell that can contract a muscle is the alpha motor neuron.
These neurons live in the ventral horn of the spinal cord (and, for the head and face, in the cranial nerve nuclei of the brainstem). Each one sends a single long axon out through the ventral root and into a muscle, where it branches and forms chemical synapses — neuromuscular junctions — onto a set of muscle fibers. When the alpha motor neuron fires, it releases acetylcholine, the fibers contract, and the body moves. One alpha motor neuron together with all the fibers it controls is called a motor unit, and a muscle is governed by a pool of such units. The brain grades the force of a contraction in two ways: by driving each active motor neuron to fire faster, and by recruiting more motor neurons into the pool. That is the whole vocabulary of muscular force — rate and recruitment — and every graceful or powerful thing you have ever done was written in it.
The English physiologist Charles Sherrington gave this arrangement its enduring name: the final common path. The phrase repays a moment’s attention, because it states the organizing fact of the entire motor system. Reflexes, central pattern generators, the deliberate plans of the cortex, the corrections of the cerebellum, the selections of the basal ganglia — all of these, however different, must in the end funnel through the same small population of alpha motor neurons, because there is no other way out. Whatever else the rest of this unit describes, it describes systems competing and cooperating for control of these cells.
This is also why the most devastating motor disease attacks precisely here. In amyotrophic lateral sclerosis (ALS, or Lou Gehrig’s disease), the alpha motor neurons degenerate. As they die, the muscles they served are cut off from every possible source of command, and the person progressively loses the ability to move — not because the plans are gone, but because the final common path is. The plans may remain perfectly intact, with nowhere to go. Damage at this level produces paralysis, the floor of the diagnostic logic we sketched in the overview, and we will return to it at the end of the chapter.
The cord is not a cable
In the overview I claimed that the spinal cord is not a cable — not a passive bundle of wires carrying the brain’s orders down and the body’s reports up, but a sophisticated controller in its own right. Here is the wiring that makes good on the claim.
Begin with the sensory side, because control requires measurement. Muscles are equipped with their own instruments: muscle spindles, which report how long a muscle is and how fast its length is changing, and Golgi tendon organs, which report how much force it is producing. The signals from these proprioceptors travel into the cord through the dorsal roots, and here is the crucial detail — they do not all rush up to the brain. The incoming fibers branch. Some do ascend toward the brainstem and cortex, carrying the information we use to know where our limbs are. But others terminate locally, right there in the cord, onto the very motor neurons that drive the muscle. This local branch is what lets sensory information modify movement almost instantly, without the long round trip to the brain and back. We build some of our control out at the periphery for the oldest of reasons: speed. Waiting for the cortex to deliberate would make us slow, and slow is dangerous.
The simplest such loop is the stretch reflex. Stretch a muscle and its spindle fires; that signal excites the muscle’s own motor neurons through a single synapse; the muscle contracts and resists the stretch. You have felt the clinical version of this, when a physician taps your patellar tendon and your leg kicks. But the everyday version is more interesting. Imagine you are holding a cup, and someone pours more liquid into it. The added weight begins to extend your arm and stretch the biceps; the spindle detects the lengthening; and before you have thought anything at all, a reflex contraction tightens the biceps and holds the cup steady. None of this went up to the brain. The brain finds out afterward, if at all. A small, fast, local controller has defended the position of your hand against a disturbance — which is, if you think about it, exactly the job of a controller.
But a joint is not run by one muscle. Joints are operated by antagonist pairs — a flexor and an extensor, biceps and triceps at the elbow — and to bend a joint you must not only contract one muscle but relax its opponent, or the two will fight and the joint will lock. (Drive both maximally at once and you can, in rare toxic states, break the bone between them.) The cord handles this automatically: the same spindle signal that excites a muscle’s own motor neurons also, through an inhibitory interneuron, suppresses the motor neurons of its antagonist. This is reciprocal inhibition — the very same motif we met in the half-centre of the overview, now wired across a joint. Coordination of opposites is built into the local circuitry.
A second interneuron is worth meeting by name, because it illustrates how much subtlety the cord contains. When an alpha motor neuron fires, it sends a side-branch onto a small inhibitory cell called a Renshaw cell, which loops back and inhibits the very motor neuron that excited it (and its neighbors). This is negative feedback in the most literal sense — the output damping itself — and it keeps contractions smooth rather than explosive, so that a reflex withdrawal is a controlled movement and not a violent jerk. We notice this circuit mainly when it fails: tetanus toxin disables inhibitory interneurons of this kind, and without their restraint, muscles contract uncontrollably into the rigid spasms that give the disease its name.
Finally, some spinal reflexes are genuinely clever, coordinating muscles across the whole body. Step on a tack and you will lift the injured foot — but if that were all you did, you would topple, because you have just removed one of your two supports. So the same reflex that flexes the injured leg simultaneously extends the other one to catch your shifting weight, recruiting muscles on both sides of the body through interneurons that cross the cord. This is the crossed-extensor reflex, and it makes the point as sharply as anything in the unit: a coordinated, whole-body, balance-preserving response is organized entirely within the spinal cord, with no help from the brain. The cord is not a cable. It is a controller that the brain inherited, sits on top of, and learns to command.
There is a complication in the stretch reflex that turns out to be one of the most elegant pieces of control engineering in the body. A muscle spindle measures length by being stretched. But what happens when the muscle contracts? As the muscle shortens, the spindle inside it goes slack — and a slack length sensor reports nothing. The instrument would fall silent at exactly the moments the system most needs it.
The solution is that the spindle has its own tiny muscle fibers, controlled by a separate population of motor neurons: the gamma motor neurons. When the brain commands a movement, it activates the alpha motor neurons that contract the big force-producing fibers and, at the same time, the gamma motor neurons that contract the little fibers inside the spindle, taking up the slack so the spindle stays taut and keeps reporting throughout the movement. This is called alpha–gamma coactivation.
Now look at what this arrangement really is. The gamma drive sets the length the spindle expects — a reference signal, a set point. The spindle then reports the difference between the muscle’s actual length and that expected length, and the stretch reflex works to drive the difference to zero. The brain, in other words, does not have to specify muscle forces directly. It can simply move the reference — “I want the muscle to be this long” — and let the local loop do the work of getting there and holding there against disturbances. That is precisely how an engineered servomechanism works, and it was sitting in your spinal cord long before anyone built one. Here, as so often in this book, the physical machinery does not implement a control law as an afterthought; the machinery is the control law.
Two ways down
If the cord is the controller, how does the brain reach it? Not through a single channel but through several descending pathways, and the way they divide the labor gives us the organizing distinction of this chapter — the one we previewed in the overview as the modern replacement for the old talk of “upper” and “lower” motor neurons. (Those terms are still useful: lower motor neurons are the alpha motor neurons of the final common path, and upper motor neurons are the cortical and brainstem neurons that drive them. The newer and more illuminating cut is not between upper and lower but between two families of descending pathway that do different jobs.)
The distinction is between distal and proximal control, and it maps onto two anatomical streams.
The lateral pathways govern distal control — the fine, independent, fractionated movements of the extremities, the hand that types and threads a needle. Chief among them is the lateral corticospinal tract, the great pathway descending from the cortex. Its axons originate in the motor cortex and neighboring areas, gather into the white-matter bundle of the internal capsule, descend through the brainstem, and then do something anatomically dramatic: about ninety percent of them cross the midline in the medulla, at a landmark whose crossing fibers form a pyramid shape — which is why the tract is also called the pyramidal tract (not, despite the coincidence, because of the pyramidal neurons that contribute to it). Having crossed, they run down the opposite side of the cord to reach the motor neurons there. The remaining ten percent descend on the same side but mostly cross over when they arrive at their target. The upshot is the feature we have met before and will meet again: the motor system is almost entirely crossed. Your left hemisphere moves your right hand. This is why a stroke in the motor regions of one hemisphere produces weakness — hemiplegia — on the opposite side of the body. (A companion pathway, the corticobulbar tract, peels off to the brainstem to control the muscles of the face, jaw, and tongue through the cranial nerves, on the same principle.)
The medial pathways govern proximal control — the trunk and the muscles near the body’s core, the machinery of posture, balance, and locomotion. These arise largely from the brainstem: the reticulospinal and vestibulospinal tracts (with smaller contributions from the tectospinal tract that turns the head and eyes toward events, and the rubrospinal tract, which is prominent in other mammals but much reduced in humans). Unlike the sharply crossed lateral pathway, these often act on both sides of the body at once — which makes sense, because keeping your balance is rarely a one-sided affair.
This anatomical split is exactly where the overview’s discussion of posture comes home. Recall the anticipatory postural adjustment: when you lift a weight on your outstretched arm, a muscle in your calf fires slightly before the arm muscle does, bracing against a loss of balance the movement is about to cause. That postural command is not coming from the motor cortex by way of the corticospinal tract. It is coming from these brainstem medial systems, launched in advance, on a prediction. The lateral pathway aims the hand; the medial pathways, quietly and largely without your awareness, keep the body that the hand is attached to from falling over. Distal control and proximal control, working at the same instant on the same act.
One more fact undercuts a tidy picture you might otherwise form. It is tempting to imagine the corticospinal tract as a clean line from the motor cortex to the muscle, but only about a third of its fibers actually originate in primary motor cortex; the rest come from premotor and supplementary areas, and even from parietal and somatosensory cortex. The “command to move,” in other words, is not issued by one place and relayed by the rest. Many regions contribute to the descending stream at once — another quiet refutation of the little man, and a hint of the distributed planning we turn to shortly.
Movements, not muscles
Now we arrive at the structure most people have in mind when they think of the motor system: the primary motor cortex, or M1, the strip of cortex (Brodmann’s area 4) lying just in front of the central sulcus, directly across that sulcus from the primary somatosensory cortex we studied earlier. Two facts about it are worth establishing before we get to the interesting controversy.
First, like the sensory cortices, M1 carries a map of the body — the motor homunculus we met in the overview, running from the throat and face low down, up through the hand, to the trunk and the foot folded over the top onto the medial wall. And, as in the sensory map, the distortions are the point: the amount of cortex devoted to a body part reflects not its size but the fineness of control we have over it. The hand and the face are enormous; the trunk and hip are small. We have exquisite control of our fingers and we talk and chew with great precision, and the map shows it.
Second, M1 looks different from sensory cortex under the microscope. The sensory areas have a thick input layer (layer 4) stuffed with cells to receive incoming signals. M1’s layer 4 is so thin it is hard to find — the cortex is called agranular for this reason — because M1 is fundamentally an output structure, not an input one. What it has instead is a powerful output layer (layer 5) containing, among others, the Betz cells: the largest neurons in the body, with thick fast-conducting axons, a minority of which run all the way to the spinal cord and synapse directly on alpha motor neurons. (Introductory accounts sometimes leave the impression that all of M1 is Betz cells driving alpha motor neurons one-to-one. It is not; that direct monosynaptic line is the exception, used for the most refined movements, not the rule.)
Now the controversy — and it is a real one, more open than you might expect for something so basic. What does activity in M1 actually represent? The naïve homunculus picture says: particular muscles. Stimulate a spot, contract a muscle. But that is not what stimulation does. Stimulate a point in M1 and you do not get a single muscle twitch; you get a small coordinated movement — the thumb flexes, the fingers curl, the mouth opens. Map it carefully and you find overlap everywhere: the same muscle can be driven from many sites, and a given site can drive a muscle as part of one movement but not another. The map is not a keyboard of muscles. It is, at best, a map of movements.
The decisive evidence came from recording single neurons in behaving monkeys, much of it pioneered by Edward Evarts. Consider a neuron that fires when the animal moves its hand in a particular direction. Now change the task so that the animal makes the same movement to the same place but, because of how it is gripping the handle, must use a different set of muscles to do it. The muscles change; the neuron fires just the same. It is signaling the goal of the movement — get the hand from here to there — not the particular muscular means of achieving it. This is the empirical heart of the slogan that organizes the modern view: M1 is about movements, not muscles.
But how can a population of broadly tuned neurons specify something as precise as a direction of reach? Here is one beautiful answer, due to Apostolos Georgopoulos. Suppose each M1 neuron has a preferred direction — it fires hardest for movements one way, less for neighboring directions, and not at all for the opposite way, a broad tuning curve rather than an on/off switch. No single neuron specifies the movement. But give each neuron a vote, weighted by how hard it is firing and pointed in its preferred direction, and add up all the votes across the population. That sum — the population vector — points reliably in the direction the limb is about to move, and you can watch it swing around in real time as the animal reaches. The movement is encoded not in any one cell but in the pattern across thousands.
This idea has had a remarkable practical afterlife. If a movement intention is legible in the population activity of M1, then in principle a computer could read that activity and act on it. This is the basis of the brain–computer interface. In a now-classic demonstration, electrodes in a monkey’s motor cortex were decoded in real time to drive a robotic arm, and the monkey learned to feed itself with the arm — guiding it by intention alone, its own hands restrained. The same approach now helps paralyzed people control cursors and prosthetic limbs, and it is the science beneath the neural-implant ventures that periodically reach the news. The link between a basic finding about cortical coding and a technology that restores movement is one of the more direct in all of neuroscience.
It would be satisfying to stop at “M1 codes movement direction,” but the question of what M1 activity fundamentally is remains genuinely unsettled, and it is worth seeing why, because the disagreement is a clean instance of this book’s central tension.
One camp emphasizes that, look closely enough, M1 activity does relate to muscles and forces, not only to abstract movement goals — the same neuron’s firing shifts when the load or the posture changes. A second camp, following Michael Graziano, used longer, more natural bouts of stimulation and evoked not twitches but complex, ethologically meaningful actions — a hand brought to the mouth, a defensive posture of the arm across the face — suggesting that M1 and nearby cortex might be organized into a map of behaviorally useful actions rather than either muscles or simple movements. This proposal is intriguing but not widely accepted, and your author would treat it with caution.
A third and increasingly influential view reframes the question entirely. Mark Churchland, Krishna Shenoy, and colleagues argue that asking what each neuron “represents” may be the wrong question — a holdover from exactly the representational thinking we warned against in Unit I. On their account, the M1 population is better understood as a dynamical system: a network whose activity, once set into the right initial state by preparation, evolves through a lawful trajectory that generates the time-varying muscle commands a movement requires. The neurons are not labels standing for directions or muscles; they are the moving parts of a pattern generator for action. Notice that this is the same shift we made for the central pattern generator in the overview — from “what does the circuit represent?” to “what dynamics does its architecture produce?” — now applied at the top of the motor hierarchy rather than the bottom. Whether the right description of M1 is muscles, movements, or dynamics is, at the time of writing, an open and active debate.
Preparing and sequencing
M1 is not the only motor area in the frontal lobe, nor the most forward-looking. Just in front of it lie two more regions that Brodmann lumped together as area 6 but that we now distinguish by what they do: the premotor cortex (on the lateral surface) and the supplementary motor area, or SMA (on the medial wall between the hemispheres). As we move forward from M1 into these areas, the activity we record becomes less about executing a movement and more about preparing, selecting, and sequencing one — more abstract, further from the muscle, closer to the plan.
The signature of premotor cortex is preparation. Give a monkey a task in which a cue tells it where to move but it must wait before moving, and many premotor neurons begin firing during the waiting period — after the cue, before the movement, holding the intention online during the delay. M1 cells, by contrast, tend to fire when the movement actually happens. Premotor activity is also notably more bilateral than M1’s sharply crossed organization. Broadly, premotor cortex seems especially involved when movements are guided by external signals — a light, a sound, the seen position of a target.
The SMA has a different specialty: sequence. Its neurons care not just about which movement is made but about the order in which movements are strung together. Record from an SMA neuron during a learned sequence — say, push, then turn, then pull — and it may fire strongly in preparation for that order; rearrange the very same movements into push, pull, turn, and the activity vanishes. The component movements are identical; only the sequence has changed, and the SMA tracks the change. Where premotor cortex leans toward externally guided action, the SMA leans toward movements generated from within — from memory and intention rather than from an external cue — and toward sequences that have been over-learned.
I can add a more direct kind of evidence here, from work I did some years ago with Itzhak Fried, recording and stimulating these areas directly in the brains of patients. Stimulating M1 produces a simple movement — a finger twitches. Stimulating the SMA produces something altogether more complex: the patient may make a coordinated, purposeful-looking movement, as though beginning some everyday act. (One patient looked as if they were about to get out of bed; another made a stereotyped two-handed gesture we took to calling the “Pope’s move.”) Most striking of all, at low stimulation intensities — too low to evoke any movement — patients sometimes reported an unprompted urge to move, a felt desire that preceded any action, and only when we raised the intensity did the movement itself appear. To stimulate a piece of cortex and have a person report the wish to act is a sobering thing, and it places the SMA close to questions about volition that we are nowhere near settling. (Damage to the SMA produces its own strange syndromes, including the alien hand, in which a patient’s hand performs purposeful actions the patient disowns — reaching out and grasping objects against their stated will.)
All of this points toward a conclusion that will carry us into the next chapter. The motor plan is abstract. Consider a demonstration that never fails to surprise. Ask someone to sign their name, and the signature is recognizably theirs. Now have them sign it with the pen taped to the wrist instead of held in the fingers; with the other hand; with the pen gripped in the teeth; even between the toes. The handwriting is clumsier, but it remains recognizably the same hand — though the person has surely never written their name with their toes before in their life. The plan cannot be a script of particular muscle commands, because the muscles are different every time and most have never performed this act. What is preserved is something higher and more general — the shape of the intended movement, its goal — with the translation into specific muscles handled late, near execution. The Russian physiologist Nikolai Bernstein, who saw this most clearly, framed the underlying difficulty as the degrees-of-freedom problem: there are vastly more ways to move the body’s many joints and muscles than there are goals to achieve, so the nervous system cannot be storing movements as fixed muscle patterns. It must work with something more abstract. We have just seen, climbing from M1 to the SMA, the cortex becoming more abstract in exactly this direction.
When the path breaks
We can now make good on the diagnostic logic promised in the overview, at least for its lower reaches. Because the motor system is layered, the kind of deficit that damage produces tells you where in the layering the damage sits — and the two failures we are equipped to understand after this chapter sit at the bottom.
Destroy the lower motor neurons, or the muscles’ connection to them, and you get flaccid paralysis: the affected muscles cannot contract at all, they lose their reflexes, and over time they waste away, because the final common path has been cut and nothing — no plan, no reflex — can reach the muscle. This is the picture in ALS, in poliomyelitis, and in injuries to the peripheral nerves. Where in the body the paralysis falls depends on which motor neurons are lost; a spinal injury high in the neck can disconnect almost the entire body below it, while a lower injury spares more.
Damage instead to the descending control — the upper motor neurons and their tracts — produces a different and revealing picture. The muscles can still contract, because their final common path is intact and other pathways still reach it; but voluntary control is weakened, and, freed from the brain’s restraining influence, the spinal reflex circuitry becomes over-active, producing spasticity and exaggerated reflexes rather than flaccid silence. The contrast between these two pictures — flaccid and areflexic below, weak and spastic above — is one of the first things a neurologist learns to read, and it is nothing more than the layered architecture of this chapter, made visible by its failures.
Notice what is not on this list. None of these disorders is a failure to know how to act, or to aim an action at a seen object. A person with these conditions, within the limits of their weakness, still understands what a key is for and still directs movements sensibly at the world. The disorders that break those capacities — the apraxias, and the inability to let vision guide the hand — sit at a higher level of the architecture, where movement meets perception and planning. That is the subject of the next chapter.
What is well established. The alpha motor neuron is the final common path: every motor command reaches muscle through it, and its loss produces flaccid paralysis. The spinal cord contains genuine control circuitry — the stretch reflex, reciprocal inhibition of antagonists, recurrent (Renshaw) inhibition, and crossed reflexes that coordinate the whole body — operating locally and fast, without the brain. Descending control divides into lateral pathways for distal/skilled movement and medial pathways for proximal/postural control, and the corticospinal system is overwhelmingly crossed, so each hemisphere governs the opposite side of the body. M1 is organized around movements more than around single muscles, and premotor and supplementary areas add preparation, external-versus-internal guidance, and the sequencing of action.
What remains contested or unsettled. What M1 activity most fundamentally reflects — particular muscles and forces, abstract movement parameters such as direction, or the state of a dynamical system that generates muscle commands — is an open and active debate. Whether M1 and nearby cortex are organized into a map of ethologically meaningful actions (the longer-stimulation findings) is intriguing but not widely accepted. The status of the “urge to move” evoked by SMA stimulation, and what it tells us about volition, is far from understood. And the degree to which the human brain has ipsilateral motor control that might be recruited after a stroke — raised in the overview — remains genuinely uncertain, with real consequences for rehabilitation.