Neuroscience: Motor Control and Whisker Sensing

Behaviour consists entirely of muscle contractions, so understanding motor control is essential for understanding any behaviour. Sensory input and movement influence each other in both directions: sensory input initiates and guides movements, and those movements in turn determine much of the sensory input the brain receives. Together they form a sensorimotor loop with no beginning or end. Neuronal computations, informed by sensory input, generate movements; these movements interact with the world and change the sensory input, which in turn influences the next neuronal computations.

Active sensing means selecting which sensory information we obtain by making specific, goal-directed movements. Most sensory information is acquired this way, in contrast to passive sensing, where a stimulus arrives without any self-generated movement, such as being touched while still.

Examples:
-In vision, head and eye movements select which regions of the visual field fall on the retina and receive attention.
-In touch, stroking an object with the fingertips generates the sensory information about its texture.
-In smell, sniffing draws more air past the nasal epithelium, increasing the number of odorant molecules detected and sampling a specific volume of air at the time of the sniff.
-Similarly, mice whisk their whiskers back and forth to scan their surroundings, so that contacts with objects generate the tactile information they need.

Mice are nocturnal animals that live in tunnels, and they use their whiskers to obtain spatial and textural information about their immediate surroundings. An exploring mouse actively moves its whiskers back and forth at ≈10 Hz, a behaviour called whisking, to scan the space around its face. When a whisker contacts an object, it bends, which drives action potentials in the sensory neurons innervating the whisker follicle. Whisking is under tight motor control, and the mouse adapts its whisking strategy to optimize the flow of sensory information for the current behaviour.

The gap-crossing task, developed by Hutson and Masterton. A mouse stands on an elevated platform in the dark, filmed under infrared light, and must locate a target platform across a gap to jump to it for a reward. The gap is so wide that the mouse can only locate the target platform by reaching across with its whiskers, palpating it before jumping. Slow-motion video shows that the whiskers on both sides mostly move together but sometimes they move out of phase, and quantifying their movements shows that each whisker is under its own individual motor control.

1. Intrinsic muscles: located entirely within the whisker pad, with one muscle for each whisker, attached at the base of that whisker's follicle and at the top of the follicle behind it. When an intrinsic muscle contracts, the whisker pivots around its insertion point in the skin and rotates forward, which is called protraction.

2. Extrinsic muscles: relatively superficial muscles anchored onto bone outside the whisker pad. When they contract, they pull the whiskers backward, which is called retraction.

=======
relate back to humans: Like flexor and extensor muscles of a limb, the intrinsic and extrinsic muscles form an opposing pair, with the same kind of neural control as all skeletal muscles (but for this unit we just consider mouseys)

1. The motor neuron's axon releases acetylcholine from synaptic vesicles at the neuromuscular junction, a chemical synapse similar to those in the CNS.

2. Acetylcholine opens nicotinic acetylcholine receptors, which are ligand-gated cation channels in the muscle membrane. This causes an an influx of cations, mainly Na+, which depolarizes the muscle fibre (aka muscle cell, myocyte, or myofibre).

3. Muscle fibres, like neurons, are excitable and have a high density of voltage-gated sodium channels, so the depolarization triggers an action potential, which spreads along the muscle fibre and into the T-tubules (which are invaginations of its membrane).

4. The depolarization from the action potential activates voltage-gated calcium channels, which interact directly with ryanodine receptors to release calcium from intracellular stores in the sarcoplasmic reticulum. The rise in cytosolic calcium allows myosin to bind to actin and pull the actin filaments past the myosin filaments, which shortens the fibre.

A motor neuron together with all of the muscle fibers it innervates, since one motor neuron can innervate many muscle cells.

By retrograde tracing with a genetically engineered rabies virus expressing a fluorescent protein (this method was developed by Wickersham and Callaway). The virus is injected into a muscle, taken up by the motor neuron terminals there, and transported backwards along the axons to the cell bodies, which become fluorescent. Injecting viruses of different colours into different muscles, such as red mCherry into the extrinsic muscles and green GFP into the intrinsic muscles, reveals where each muscle's motor neurons are located in brainstem sections (e.g. only the motor neurons that innervate intrinsic muscles will become green).

Rabies virus naturally spreads retrogradely across synapses, from neurons to their presynaptic partners. In the engineered version, called ΔG rabies, the gene for the glycoprotein G, which the virus needs to cross synapses, is deleted and replaced with a gene for a fluorescent protein. Used alone, ΔG rabies infects neurons whose axon terminals are at the injection site and labels them, but it cannot spread any further. If the initially infected neurons are made to express G, then the virus can cross exactly one synapse back into their presynaptic partners. Those partners do not express G, so the spread stops there, so this method labels only the neurons directly connected to the starting cells.

The motor neurons that all innervate a given muscle are clustered close together in a tightly segregated region called a motor pool, an organizing principle found throughout the brainstem and spinal cord. The whisker motor neurons are cholinergic neurons in the facial nucleus of the brainstem, which contains all the motor neurons controlling facial movements. All facial motor neurons (including whisker motor neurons) lie on the same side as the muscles they innervate (ipsilateral). The motor pools for the intrinsic (protraction) and extrinsic (retraction) muscles lie in slightly different locations within the facial nucleus.

Premotor neurons are the neurons that make synapses onto motor neurons, providing the excitatory glutamatergic and inhibitory GABAergic and glycinergic inputs that control motor neuron firing. They can be labelled with ΔG rabies virus: the virus infects the motor neurons, which are made to express the glycoprotein G, so it crosses exactly one synapse back and labels only the neurons connected directly onto the motor neurons. For the whisker muscles, as for all other muscles, the premotor neurons are found in many different brain regions rather than in one place, but there are hotspots where many premotor neurons cluster together.

One hotspot of whisker premotor neurons is the brainstem immediately behind and around the facial nucleus. There, protraction and retraction premotor neurons are partly intermingled, but retraction premotor neurons are concentrated in the spinal trigeminal nucleus, while protraction premotor neurons are concentrated in the reticular formation. Consistent with this, electrical microstimulation of the spinal trigeminal nucleus causes whisker retraction, while stimulation of the reticular formation causes whisker protraction.

Local circuits in the brainstem and spinal cord generate reflexes and simple sensorimotor interactions, because sensory information from the whiskers and other senses reaches premotor neurons there directly. By contrast, the cortex is thought to drive voluntary movement, but mostly indirectly: cortical neurons rarely connect strongly to motor neurons themselves, and instead act through brainstem and spinal cord circuits that control the premotor neurons. So to understand voluntary control of whisker movements, we need to study the cortical inputs to the premotor neurons (e.g. inputs from the whisker sensory cortex and whisker motor cortex to the brainstem)

Labelling their axons with an anterograde viral tracer shows two distinct targets:

1. Whisker sensory cortex projects mainly to the interpolaris region of the spinal trigeminal nucleus, where the retraction premotor neurons are located. Activating it with channelrhodopsin-2 retracts the whiskers on the opposite side of the face, and they stay retracted for as long as the light is on.

2. Whisker motor cortex projects mainly to the reticular formation, where the protraction premotor neurons are located. Activating it causes rhythmic whisking, resembling the exploratory whisking seen in the gap-crossing task. The whisker is repeatedly protracted and returns backward in between. Direct electrical stimulation of the reticular formation, as seen in the previous card, simply protracts the whisker. Cortical input, however, probably recruits circuits there that act as central pattern generators, driving the protraction premotor neurons rhythmically

Facial motor neurons control the whiskers on their own side (ipsilateral): the right facial nucleus moves the right whiskers. Each cortical hemisphere, by contrast, controls the whiskers on the opposite side (contralateral), because its descending projections cross to the brainstem on the other side. Stimulating the left cortex therefore drives premotor and motor neurons in the right brainstem, which move the right whiskers. Whisker sensory cortex also receives touch information from the opposite side's whiskers, so its retraction signal acts on the same whiskers it senses.

1. Sensory cortex, by retracting the whisker via premotor neurons in the interpolaris region of the spinal trigeminal nucleus, may provide a negative feedback signal that stops a movement after contact. Retraction reduces whisker-object contact and bending, producing smaller, possibly more refined sensory signals, like stopping your hand when it touches a wall while exploring a dark room.

2. Motor cortex, by driving rhythmic whisking via central pattern generator circuits in the reticular formation, which drive the protraction premotor neurons, may act to gather more sensory information by moving the whiskers forward to explore. Moving the whiskers more makes them touch objects more often, generating more sensory input.

=====
So M1 drives exploration to gather sensory information, while S1 stops the movement once contact is made

On trials where the mouse starts exploratory whisking, S1 responds strongly, as if the mouse clearly registered the touch, and the activity then spreads to motor cortex, probably through a direct projection from S1. Because S1 drives whisker retraction, this strong S1 response may first cause a small retraction of the whisker. Motor cortex activity may then drive exploratory whisking, as the mouse searches for whatever touched its whisker. On trials where the mouse ignores the stimulus, S1 responds more weakly and motor cortex is not activated. These results show a correlation between cortical activity and behaviour, not that the activity causes the behaviour.

=====
Note: this is an example of passive touch because the whisker was deflected while the mouse sat still, without any movement of its own. The whisking afterward was the mouse's response to that touch, not what produced it.

The variability probably reflects context-dependent processing. The state of the cortex when the stimulus arrives, such as its ongoing activity and how alert or engaged the mouse is, affects how strongly S1 responds and whether the signal reaches motor cortex, and therefore whether the mouse responds with exploratory whisking or ignores the stimulus.

In whole-cell recordings while the mouse whisks against a pole placed in its whiskers' path (active touch since the mouse is actively whisking rather than passively being touched), each touch evokes a clear depolarization in many pyramidal neurons (i.e. the touches are correlated with depolarizations). This depolarization consists of glutamatergic EPSPs arriving through the ascending sensory pathway, and it sometimes triggers action potentials. Each individual touch is therefore encoded in primary somatosensory cortex. To determine where the object is, however, the mouse must combine the timing of each touch with the position of its whisker, thereby integrating sensory and motor signals.

They are directly and reciprocally connected by long-range axons of excitatory pyramidal neurons. Motor cortex axons innervate the sensory cortex in two zones: superficially in layer 1, where they contact the distal dendrites of excitatory neurons, and deep in layers 5 and 6.

Somatostatin-expressing GABAergic neurons (Martinotti cells) act as a gate, so that motor cortex input is strong only while the mouse is whisking. Motor cortex axons arrive in layer 1 of barrel cortex, on the distal dendrites of excitatory neurons, and Sst neurons inhibit these same dendrites. When the mouse sits still, Sst neurons fire at relatively high rates, so the dendrites are inhibited and motor input has little effect. When the mouse starts whisking, Sst neurons hyperpolarize and stop firing, so the inhibition is lifted and motor input can strongly influence the excitatory neurons.

By letting motor cortex input through only while the mouse whisks, this gating could allow barrel cortex neurons to integrate motor signals about whisker position with the sensory input they also receive, for example to determine where the touched object is.

1. Through movement: we actively gather much of our sensory information with our own movements, for example moving our eyes and head to choose what we see, or moving our fingers over an object to feel it.

2. Through the brain's own activity: the brain actively builds what we perceive from the sensory input, shaped by our past experience and the current context, so the same input can be perceived differently by different people or under different circumstances. This construction is thought to happen largely in the neocortex.

1. Bistable images: images that can be perceived in two different ways, with perception spontaneously alternating between the two even though the image itself does not change (e.g. face/vase image). Since the same light falls on the retina in both cases, percepts must be internal constructs generated by neuronal activity, not direct copies of the world.

2. Object recognition: the ability to identify\/recognize what we are looking at, such as recognizing a silhouette as a human face. We can do this only because we have seen many related images before, and the brain has extracted the statistics linking them. This learning begins in infancy, when babies gradually learn to perceive the world

Percepts are subjective, so a subject must be able to report what it perceives. A causal, mechanistic understanding requires measuring the activity of individual neurons and how they communicate, which currently cannot be done in humans. Animals must therefore be trained to report their percepts, while their neuronal activity is measured and perturbed during the perceptual task.

A well-controlled sensory stimulus is delivered, evoking neuronal computations that depend on context, learning and the subject's current goals. The subject then reports its percept through a motor output: for example, a human viewing a bistable image could press one button when seeing the faces and another when seeing the cup. Perception can therefore be studied as a learned, goal-directed sensorimotor transformation, which may be the minimal essential core of sensory perception.

Detection, i.e. deciding whether a stimulus occurred or not. A head-restrained mouse sits in a cardboard tube, with a plate restraining its forepaws so it cannot touch its face. A small piece of metal is attached to one whisker, and an electromagnetic coil beneath the mouse delivers a magnetic pulse lasting ≈1 ms, deflecting the whisker at random times with no warning cue. A thirsty mouse learns that a whisker deflection means water is available: if it licks the spout within a 1 s window after the stimulus, a sensor on the spout detects the lick and a valve delivers a drop of water. The lick therefore serves as the mouse's report that it perceived the stimulus, turning the percept into a measurable motor output.

1. Stimulus trials, where the deflection is delivered:
-Hit: the mouse licks within the reward window and is rewarded, indicating that it detected the stimulus.
-Miss: the mouse does not lick, interpreted as not having perceived the stimulus.
The hit rate is the fraction of stimulus trials that are hits.

2. Catch trials, with no stimulus delivered:
-False alarm: the mouse licks anyway, and no reward is given.
-Correct rejection: the mouse does not lick.
The false alarm rate is the fraction of catch trials with licking. It measures spontaneous licking, providing a baseline for judging whether licking on stimulus trials is really driven by the stimulus or just spontaneous.

On the first day, the mouse knows that licking is sometimes rewarded but not when, so it licks equally often with and without a stimulus, and the hit rate and false alarm rate are similar. Through trial-and-error, reward-based learning, it learns that licking after a stimulus is rewarded, so its hit rate increases, eventually to ≈70 to 80% of stimuli. It also learns to stop licking when there is no stimulus, so its false alarm rate decreases. The increasing difference between the hit rate and the false alarm rate shows that the task has been learned (see image).

This is shown by a psychometric curve, which plots stimulus strength against behavioural performance. As the whisker deflection is made weaker (i.e. as we reduce stimulus strength), the hit rate decreases. The psychophysical threshold is the stimulus strength at which the hit rate (performance rate) is half of its maximum. Below a certain strength, the mouse no longer licks more often than its false alarm rate, so the stimulus is undetectable and any licking is just spontaneous.

Drugs are injected into the C2 barrel column, the part of S1 representing the stimulated whisker:
1. Tetrodotoxin blocks voltage-gated sodium channels, preventing action potential firing.
2. CNQX and APV block AMPA and NMDA receptors, preventing glutamatergic synaptic transmission.

In the experiment, the C2 whisker was stimulated and the C2 barrel column inactivated. Tetrodotoxin causes a rapid, dramatic drop in performance, with the hit rate falling from ≈0.85 to ≈0.3, and CNQX with APV causes a similar deficit. Performance recovers as the drugs wear off. As controls, injecting only the solution the drugs are dissolved in (Ringer's solution) has no effect, and neither does tetrodotoxin injected into the forepaw area of S1, ≈1 mm away. The deficit is therefore specific to silencing the barrel column representing the stimulated whisker, showing that action potential firing and glutamatergic synaptic transmission there are necessary for the detection task.

A virus is used to express channelrhodopsin-2 in the excitatory neurons of the C2 barrel column, the barrel column of the trained whisker. This takes a few weeks, during which the mouse is trained on the normal whisker detection task with the C2 whisker, without any light. On a test day, the whisker stimulus is replaced by a 5 ms blue light flash on the C2 barrel column. In most mice, the very first flash triggered licking, so the mouse responded to the cortical stimulation as it would to a whisker stimulus. This shows that activating the excitatory neurons of the barrel column representing the trained whisker is sufficient to drive the learned response (i.e. the peripheral stimulus can be replaced by stimulation of neocortical neurons).

In the reversed experiment, mice are trained to lick in response to 5 ms blue light flashes on the C2 barrel column, rather than to whisker stimuli. When their C2 whisker is then deflected for the first time, they lick in response to the very first stimuli. The transfer therefore works in both directions, so behaviour can effectively be programmed optogenetically from S1, and the mice appear to treat cortical stimulation and whisker stimulation as equivalent.

Silencing the barrel column representing the stimulated whisker abolishes detection, so its activity is necessary (necessary and sufficient for the behavioural report i.e. the licking: necessary for the whisker stimulus to produce licking, and sufficient to produce licking on its own, without any whisker stimulus). Activating that barrel column optogenetically triggers the same learned licking as a whisker stimulus, so its activity is also sufficient. However, S1 itself does not produce the licking: downstream brain areas presumably convert the S1 activity representing the touch into the motor command for licking to obtain the reward.

Averaging the membrane potential separately over all hit trials and all miss trials reveals two components:

1. Early response: shortly after the whisker deflection, the pyramidal neurons depolarize by the same amount on hit and miss trials. This depolarization is presumably driven directly by sensory input arriving from the thalamus. Because the stimulus, and therefore this input, is identical on every trial, S1 reliably signals that the whisker was deflected, regardless of whether or not the mouse goes on to lick.

2. Late response: from ≈200 to 400 ms after the stimulus, the pyramidal neurons are more depolarized and fire more action potentials on hit trials than on miss trials. This late activity differs depending on whether the mouse licks, so it correlates with the mouse's percept (i.e. it is linked to when the mouse detects a stimulus and licks). It also occurs hundreds of milliseconds before most licks, so it cannot be a consequence of licking and could instead help cause the percept and the response.

Stanislas Dehaene and colleagues recorded EEG over visual cortex while presenting visual stimuli close to the threshold of conscious perception, so that some stimuli were perceived and others were not. Both perceived and unperceived stimuli evoked a similar early, reliable response. Later activity differed, correlating with whether the stimulus was consciously perceived, i.e. late activity was larger for perceived stimuli. In humans, however, it is difficult to test whether this late activity actually causes perception.

S1 activity must be silenced specifically during the late period, which requires inactivation with high temporal precision. Channelrhodopsin-2 is expressed specifically in parvalbumin-expressing GABAergic neurons, which inhibit pyramidal neurons near the soma. Blue light makes these neurons fire, releasing GABA and hyperpolarizing the surrounding excitatory neurons, which shuts down S1 activity only while the light is on. The light can therefore be timed to silence either the early sensory response or only the late depolarization.

1. Silencing S1 for a brief ≈50 ms window around the time of the stimulus abolishes the early sensory response and impairs detection, matching the pharmacological inactivation but on a much shorter timescale. The early response, which occurs on every trial, registers the stimulus in S1 and is therefore necessary for the stimulus to be detected.

2. Silencing only later, ≈100 to 300 ms after the stimulus, leaves the early sensory response intact but removes the late depolarization, and still causes a small but significant deficit in detection. The late depolarization, which is stronger on hit trials, therefore not only correlates with whether the mouse perceives and reports the stimulus but also causally contributes to it, similar to the late activity linked to conscious perception in humans.

There are many types of learning, including declarative, procedural, fear, spatial and motor learning, which may involve changes in different neuronal circuits in different brain regions. The hippocampus and neocortex contribute to many of them, for example spatial learning through NMDA receptors in hippocampal CA1. Reward-based learning, driven by the motivation to obtain rewards, may be more ancient and fundamental, likely existing long before the neocortex evolved.

Actions that consistently lead to reward should be repeated, but the rewarded action often depends on the context, which is provided by sensory input. The animal must therefore learn that in a specific sensory context, a specific motor output leads to reward. In a network of synaptically connected neurons, this could be achieved by strengthening the most direct pathway linking the neurons activated by the sensory input to those driving the rewarded motor output, for example through LTP of the glutamatergic synapses along it. Reward signals should therefore drive synaptic plasticity in the brain's sensorimotor networks to ensure that the relevant sensory input reliably drives the rewarded motor output.

The reward arrives only after the action, when the neuronal activity that produced the action is already over. To reinforce the right behaviour, the reward signal must therefore strengthen the recently active circuits that underlie the rewarded sensorimotor transformation, rather than whatever is active when the reward arrives. A prominent hypothesis is that important aspects of reward-based learning occur through synaptic plasticity in the striatum, controlled by dopamine reward signals.

By midbrain dopaminergic neurons, in both the VTA and SNc. These neurons fire tonically at a baseline rate of ≈5 Hz. When an unexpected reward is delivered, such as juice given to a thirsty animal, their firing increases sharply ≈100 ms later, in a brief phasic burst lasting ≈100 ms.

This was first shown in monkeys by Wolfram Schultz and confirmed in mice by Naoshige Uchida, where the recorded cells could also be verified as dopaminergic through optogenetic tagging: channelrhodopsin-2 was expressed only in dopaminergic neurons, so cells that fired immediately in response to light flashes could be identified as dopaminergic.

Dopaminergic neurons signal the reward prediction error: the difference between the reward received and the reward predicted, rather than simply signalling reward itself.

1. Before learning: the reward is unexpected, so it is better than predicted, and the neurons fire a phasic burst above their tonic baseline when it arrives.
2. After learning, through classical conditioning, that a cue such as a sound or light predicts the reward ≈1 s later: the reward is now fully expected, so firing does not change when it arrives. Instead, the burst occurs at the cue, the earliest predictor of reward.
3. After learning, if the expected reward is omitted: the outcome is worse than predicted, so firing drops below the tonic baseline at the time the reward was expected.

This error signal is thought to act as a teaching signal in many theories of reward-based learning, strengthening behaviours that lead to more reward than predicted.

They project most strongly to the striatum, which is divided into the dorsal striatum and the ventral striatum aka nucleus accumbens. Dopaminergic neurons form two groups:

1. The ventral tegmental area (VTA), which heavily innervates the nucleus accumbens and also projects to PFC. These mesolimbic and mesocortical pathways are associated mainly with reward and motivation.

2. The substantia nigra pars compacta (SNc), which projects to the dorsal striatum. This nigrostriatal pathway is associated mainly with movement and action selection, and its degeneration causes Parkinson's disease.

Dopamine is released from vesicle-filled swellings along the axons that have no closely apposed postsynaptic membranes, so it diffuses through the extracellular space via volume transmission to reach metabotropic receptors.

Both are G-protein-coupled receptors (metabotropic receptors) with seven transmembrane domains.

1. D1 receptors couple to the G proteins Gs and Golf, which stimulate adenylate cyclase and increase production of the second messenger cyclic AMP.

2. D2 receptors couple to Gi and Go, which have several targets, including phospholipase C, and they also inhibit adenylate cyclase, decreasing cyclic AMP.

The two receptors therefore have partly opposing effects, and they are expressed on two distinct types of striatal projection neuron.

In mice expressing Cre-recombinase under the D1 or the D2 receptor promoter, a virus carrying a Cre-dependent (floxed) GFP construct is injected into the striatum. GFP is therefore expressed only in the D1- or D2-expressing neurons, so the axons of only one neuron type are labelled.

1. D1-expressing neurons project to the substantia nigra pars reticulata (SNr), right next to the dopaminergic neurons of the SNc that project back to the striatum.

2. D2-expressing neurons project to the external part of the globus pallidus, which is much closer to the striatum than the SN.

All striatal projection neurons are GABAergic, and the two pathways have opposing effects on the substantia nigra pars reticulata (SNr), whose tonically active GABAergic neurons inhibit brainstem motor centres and motor thalamus:

1. Direct (striatonigral) pathway: D1-expressing neurons inhibit the SNr directly, decreasing its activity. This releases the motor centres from inhibition, so it acts as a go pathway.

2. Indirect (striatopallidal) pathway: D2-expressing neurons inhibit the external globus pallidus, whose GABAergic neurons normally inhibit the SNr. This disinhibits the SNr, increasing its activity and strengthening its inhibition of the motor centres, so it acts as a no-go pathway.

Because the two pathways express different dopamine receptors and have opposing effects on movement, dopamine reward signals can act on them in opposite ways to favour rewarded actions.

At corticostriatal synapses, which are part of the pathway that generates the animal's behaviour. Cortical pyramidal neurons carry the sensory context to the striatum, releasing glutamate onto direct and indirect pathway projection neurons, whose output then helps trigger a motor action, such as licking in the whisker detection task. These same projection neurons express D1 or D2 receptors. When that action is rewarded unexpectedly, the phasic dopamine burst reaches the striatum just after these corticostriatal synapses have been active, so it can strengthen or weaken the specific connections that produced the rewarded action.

1. Through D1 receptors, dopamine promotes the insertion of AMPA and NMDA receptors into the postsynaptic membrane, and D1 activation appears essential for LTP at cortical synapses onto D1-expressing striatal neurons. Cortical input onto direct pathway neurons is therefore strengthened when an unexpected reward is received, increasing the go signal that promotes the rewarded action.

2. Through D2 receptors, dopamine prevents calcium entry through NMDA receptors and promotes LTD at cortical synapses onto D2-expressing striatal neurons. Cortical input onto indirect pathway neurons is therefore weakened when an unexpected reward is received, which reduces the no-go signal that would otherwise suppress the rewarded action.

So the effects are complementary: dopamine increases direct pathway activity and decreases indirect path activity, with both changes favouring the rewarded action

The SNr contains tonically active GABAergic neurons that inhibit brainstem motor centres and motor thalamus, preventing unwanted movements. During learning, a sensory context, such as a whisker stimulus, activates cortical inputs to the striatum. If the resulting action is rewarded, dopamine then changes the strength of those inputs. When the same context occurs again:
1. The strengthened input onto direct pathway neurons makes them inhibit the SNr more, releasing the motor centres from inhibition and promoting the rewarded movement.
2. The weakened input onto indirect pathway neurons makes them inhibit the external globus pallidus less. The globus pallidus then inhibits the SNr more strongly, also promoting the movement.

These changes could therefore form a synaptic mechanism for reward-based learning.

Direct pathway: cortex → D1 striatal neurons --| GPi/SNr --| thalamus → cortex. Net effect: the thalamus is disinhibited, promoting movement.

Indirect pathway: cortex → D2 striatal neurons --| GPe --| STN → GPi/SNr --| thalamus → cortex. Net effect: the thalamus is inhibited more, suppressing movement. GPi/SNr also --| brainstem motor areas, with the same effect on movement.

The whisker deflection excites S1 through the thalamus, and S1, which is necessary for the task, sends glutamatergic projections to the striatum. On hit trials, the water reward should cause a burst of dopamine firing at the time these corticostriatal inputs have just been active. This could potentiate cortical input onto direct pathway neurons, strengthening the go pathway, and depress cortical input onto indirect pathway neurons, weakening the no-go pathway. After learning, the whisker stimulus would then activate direct pathway neurons, which inhibit the SNr, releasing the brainstem motor nuclei to drive licking.

Membrane potential recordings show that the whisker stimulus evokes a brief, transient depolarization in direct pathway (D1-expressing, dSPN in image) striatal projection neurons, driven by EPSPs from their glutamatergic inputs. This depolarization is missing or much weaker in indirect pathway (D2-expressing, iSPN in image) neurons. This transient excitation of the direct pathway could act as a go signal that initiates licking.

After training on the whisker task, channelrhodopsin-2 expressed in direct pathway neurons is activated with a brief, few-millisecond blue light flash. This drives licking even more reliably than the whisker stimulus, with low false alarm rates, so direct pathway activity is sufficient to trigger the learned, goal-directed licking response. Stimulating indirect pathway neurons does not drive licking and, if anything, prevents it. The two pathways therefore have opposing functions, consistent with the hypothesis that reward-based learning strengthens the direct (go) pathway and weakens the indirect (no-go) pathway to favour rewarded actions.

According to Gustavsson et al. (2011), who studied 30 European countries with a combined population of over 500 million, brain disorders in 2010 affected over 150 million people, more than a quarter of the population. They cost ≈800 billion euros (≈1 trillion US dollars), over 5% of Europe's total GDP of ≈14 trillion euros. About 40% of this cost was direct medical costs, 20% direct non-medical costs and 40% mainly lost working hours. Estimates for the United States are similar or slightly higher, and with an ageing population, both the number of people affected and the costs are expected to rise.

Of 19 categories of brain disorders, the five most costly in 2010 were:
1. Mood disorders, mainly depression: ≈33 million people, ≈113 billion euros
2. Dementia, including Alzheimer's disease: ≈6 million people, ≈105 billion euros
3. Psychotic disorders, including schizophrenia: ≈5 million people, ≈94 billion euros
4. Anxiety disorders, including panic disorder and post-traumatic stress disorder: ≈69 million people, ≈74 billion euros
5. Addiction, mainly drug and alcohol abuse: ≈16 million people, ≈66 billion euros

with Parkinson's being 15th.

Parkinson's disease is an age-related, slow, progressive neurodegenerative disorder, first described by James Parkinson in his 1817 "Essay on the Shaking Palsy." It affects ≈1.2 million people in Europe, about 1 in 500, at a cost of ≈14 billion euros. It's cardinal motor symptoms include:
1. Bradykinesia: slowness of movement and difficulty initiating movements
2. Problems with walking and posture
3. Tremor: the shaking that Parkinson described
4. Rigidity: unwanted contraction of muscles

These worsen as the disease progresses, and at later stages, nonmotor symptoms like dementia and depression can become prominent.

Degeneration of the dopaminergic neurons of the substantia nigra pars compacta (SNc), which project to the striatum. The dopaminergic neurons of the VTA, projecting to the NAc and PFC, are less affected, so the degeneration is somewhat specific to SNc neurons. The degenerating neurons contain Lewy bodies, which are cytoplasmic, protein-rich inclusions with high concentrations of the protein alpha-synuclein, whose aggregates may cause the degeneration. By the time motor symptoms are diagnosed, ≈60% of these dopaminergic neurons have already died, and striatal dopamine is reduced by ≈80%.

It may not begin in the substantia nigra pars compacta, but rather in the brainstem or in the periphery (e.g. the gut) with alpha-synuclein possibly spreading along synaptic pathways in a prion-like process. By the end stage, Lewy bodies and degenerating neurons are found throughout the neocortex, so much of the brain is affected. Because so many dopaminergic neurons are lost before motor symptoms appear, earlier nonmotor symptoms presumably exist but are hard to detect, and earlier diagnosis is an important goal for future research.

Some people who injected illicit drugs contaminated with MPTP (methyl-phenyl-tetrahydropyridine) rapidly developed Parkinsonian symptoms. MPTP is lipophilic, so it crosses the blood-brain barrier, and in the brain it is metabolized to the cation MPP+ (1-methyl-4-phenylpyridinium). MPP+ is toxic to mitochondria and causes dopaminergic neurons to degenerate. Mitochondrial dysfunction is also implicated in the degeneration of these neurons in Parkinson's disease more broadly, not only after MPTP exposure. MPTP produces the same dopaminergic neuron loss and motor symptoms in animals, so it is used as an animal model of Parkinson's disease (i.e. mimic PD by administering MPTP to the animal) to test possible treatments.

Anything that causes degeneration of the dopaminergic neurons of the substantia nigra pars compacta is likely to produce similar motor symptoms. Pesticides and insecticides are recognized risk factors, and Parkinson's disease is often more prevalent in rural areas. However, not everyone exposed to them develops the disease, so it probably results from an interaction between genes and environment.

Two key genes implicated in Parkinson's disease are alpha-synuclein and LRRK2. Between 5% and perhaps 50% of cases have a genetic element. Genome-wide association studies, which correlate the prevalence of the disease with genetic variants across the population, have identified many susceptibility loci: regions of the genome, containing genes or their regulatory DNA, where mutations are associated with an increased risk. However, the normal physiological functions of alpha-synuclein and LRRK2 are unknown, so much basic research is needed to understand how their mutations lead to the disease.

1. Alpha-synuclein, the main protein in Lewy bodies: many independent mutations have been found, varying in how strongly they increase risk:
-A53T, a mutation in which alanine at amino acid 53 of the protein is replaced by threonine, is autosomal dominant: a single copy makes Parkinson's disease very likely.
-REP1 is a dinucleotide repeat expansion in a regulatory region upstream of the gene. It has a much smaller effect, slightly increasing the risk or changing the age of onset.

2. LRRK2 (leucine-rich repeat kinase 2): many variants have also been found, such as R1441C/G/H, I2020T, Y1699C and G2019S. They may cause disease through abnormally increased kinase activity.

L-DOPA. Because the disease reduces brain dopamine, the aim is to restore it, but dopamine itself cannot cross the blood-brain barrier. L-DOPA, the natural precursor of dopamine, is taken as a pill, absorbed into the bloodstream and carried to the brain, where it crosses the blood-brain barrier. Inside the brain, the enzyme DOPA decarboxylase converts it into dopamine, greatly relieving the symptoms. Its effectiveness declines over time, however: as more dopaminergic neurons die, fewer terminals remain to store dopamine and release it steadily, so dopamine levels rise and fall sharply with each dose. These swings contribute to serious side effects at high doses and with long-term use, including dyskinesia, unwanted movements, and sometimes changes in character.

(L-DOPA treatment was developed in the 1950s by Arvid Carlsson, who received the Nobel Prize in 2000 for this work and for identifying dopamine as a neurotransmitter).

Deep brain stimulation (DBS), in which electrodes are implanted into the brain to stimulate the tissue at their tips. For Parkinson's disease, electrodes are bilaterally implanted in the subthalamic nucleus (i.e on both sides). Because this structure is small, the surgery is difficult, and a neurophysiologist helps the neurosurgeon guide the electrodes. After recovery, high-frequency stimulation at 100 Hz (one pulse every 10 ms) is applied continuously for the rest of the patient's life, giving almost immediate relief of symptoms. Because DBS acts on the basal ganglia circuit rather than depending on surviving dopaminergic terminals, its effect is steady, and it usually allows L-DOPA doses to be reduced, lessening the fluctuations and dyskinesia.

1. Striatum: made up of the dorsal striatum (divided into the caudate nucleus and the putamen by the axons of the internal capsule) and the ventral striatum (nucleus accumbens). All parts contain mainly GABAergic projection neurons (medium spiny neurons). The striatum is the main input structure of the basal ganglia and projects to the globus pallidus and substantia nigra.
2. Globus pallidus: an external segment and an internal segment, both made of GABAergic projection neurons. The external segment receives the indirect pathway and inhibits other basal ganglia nuclei, while the internal segment is an output nucleus that inhibits the thalamus.
3. Substantia nigra: the pars reticulata contains tonically active GABAergic neurons and, like the internal globus pallidus, is an output nucleus, inhibiting the thalamus and brainstem motor areas. The pars compacta contains the dopaminergic neurons projecting to the striatum.
4. Subthalamic nucleus: the only excitatory part of the basal ganglia, made of glutamatergic neurons that provide excitatory drive to other basal ganglia structures.

Dopamine normally favours movement in two ways: through D1 receptors, it strengthens cortical input onto the direct pathway, which inhibits the basal ganglia output nuclei (the SNr and GPi), and through D2 receptors, it weakens cortical input onto the indirect pathway. Without dopamine, both effects are lost: the direct pathway weakens, and the indirect pathway is no longer held in check, so it may become overactive. Both changes make the output nuclei more active. The output nuclei then inhibit the thalamus more strongly, which reduces motor output from the cortex. They also inhibit brainstem motor areas more strongly through their descending projections. The overall decrease in movement matches the slowness and difficulty initiating movements in Parkinson's disease.

The STN contains glutamatergic neurons that excite the GPi and SNr. If stimulation simply made subthalamic neurons fire, it would excite these GABAergic output nuclei further, increasing their inhibition of the thalamus and brainstem motor areas and worsening the movement deficit. Since stimulation actually relieves the symptoms, it cannot work simply by exciting the STN.

1. Inactivation: 100 Hz stimulation may silence the subthalamic nucleus rather than excite it, reducing excitation of the basal ganglia output nuclei and allowing movement.
2. Axon stimulation: the benefit may come from stimulating axons that terminate in the region, such as those of cortical neurons, globus pallidus neurons or dopaminergic neurons.
3. Disrupting abnormal rhythms: Parkinson's patients show prominent beta oscillations, at ≈10 to 20 Hz, which stimulation may disrupt.

The mechanism remains largely unknown, and understanding it could lead to more specific therapies with fewer side effects.
Recall the approximate values for:
-Whisker sensorimotor control: whisking frequency, magnetic stimulus duration
-Whisker detection task: reward window, learned hit rate, hit rate after S1 inactivation, distance of control injection, optogenetic flash duration, timing of early and late S1 silencing
-Dopamine signalling: baseline firing rate, burst latency and duration, cue-to-reward interval
-Brain disorders in Europe: population studied, people affected, total cost and share of GDP, cost breakdown
-Parkinson's disease: prevalence, cost, neuron and dopamine loss at diagnosis, genetic contribution

-Whisker sensorimotor control: whisking ≈10 Hz; magnetic whisker pulse ≈1 ms
-Whisker detection task: 1 s reward window; trained hit rate ≈70 to 80%; S1 inactivation drops hit rate from ≈0.85 to ≈0.3; control injection in forepaw S1 ≈1 mm away; optogenetic flashes 5 ms; early silencing ≈50 ms around the stimulus, late silencing ≈100 to 300 ms after; late depolarization precedes licking by hundreds of ms
-Dopamine signalling: baseline ≈5 Hz; burst ≈100 ms after unexpected reward, lasting ≈100 ms; predictive cue ≈1 s before reward
-Brain disorders in Europe (2010): >500 million people in 30 countries; >150 million affected; ≈800 billion euros (≈1 trillion US dollars), >5% of GDP; ≈40% direct medical, ≈20% direct non-medical, ≈40% lost work
-Parkinson's disease: ≈1.2 million people in Europe (≈1 in 500), ≈14 billion euros; at diagnosis ≈60% of SNc dopaminergic neurons lost and striatal dopamine reduced by ≈80%; ≈5 to 50% of cases have a genetic element; A53T mutation at amino acid 53
-Deep brain stimulation: 100 Hz (one pulse every 10 ms), applied continuously; >100,000 patients implanted; abnormal beta oscillations ≈10 to 20 Hz