Neuronal Plasticity CHAPTER
3
CONTENTS FEATURES
3.1 How Are Synapses Strengthened in the Marine Snail EVOLUTION IN ACTION
Aplysia?
Box 3.1 The Impact of Invertebrates
3.2 How Are Synapses Strengthened in Mammals? on Neurobiology
3.3 When Are Synapses Weakened? A NEUROBIOLOGY IN DEPTH
Dorsal hippocampus
Box 3.2 H ippocampalSSptirnuacltcuorredand Functions
3.4 Can Inactive Neurons Strengthen Their Inputs?
THERAPIES
3.5 Can Experiences Rewire the Brain?
Box 3.3 B rain–Machine Interfaces
3.6 How Does Experience Affect Brain and Cortex Size?
Ventral
A hippocampus
3.7 Does Neural Plasticity CauHseeadLearning and MOlfeamctooryrybu?lb
B
1 cm
B
Giant
cell body
Hippocampus 71
Temporal lobe
72 Chapter 3 Neuronal Plasticity
Computers can store vast amounts of information by f lipping tiny electronic
switches (transistors) on or off, but they cannot change their own, internal wiring.
Brains, in contrast, are plastic. They tend not to add more neurons once the organism
has reached adulthood, but the connections between existing neurons can and typic-
ally do change as a result of experience. Good evidence for neuronal plasticity has
accumulated over the last 50 years or so, but the idea itself is old. James and Meynert
(see Chapter 1) had relied on the notion of neuronal plasticity to explain how a child
learns not to touch a flame. They had proposed that mental associations (between the
sight of a f lame, pain in the finger, and withdrawal of the arm) result from the
strengthening of connections between groups of cortical neurons that represent
those percepts and movements. A more detailed vision of neuronal plasticity was
proposed by Ramón y Cajal, who wrote
One can admit as highly probable that mental exercise promotes in the most involved
areas a greater development of the protoplasmic extensions [i.e., dendrites] and [axon]
collaterals. As a result, associations that have already been created between certain cell
groups strengthen themselves, notably by multiplying the terminal twigs of the proto-
plasmic extensions [i.e., dendrites] and [axon] collaterals. In addition, entirely new
intercellular connections may establish themselves thanks to the formation of new
collaterals and the expansion of [dendrites].
Ramón y Cajal, 1894, pp. 466–467, author’s translation.
Thus, the notion that brains can reorganize themselves as a result of “mental exer-
cise” is hardly new. Yet direct evidence for neuronal plasticity did not accumulate
until the second half of the twentieth century, when new methods made it possible to
generate detailed maps of the brain’s functional organization, trace neuronal connec-
tions at microscopic scales, and quantify the strength of synapses.
3.1 How Are Synapses Strengthened
intheMarineSnail Aplysia?
How the brain rewires itself as a result of “mental exercise” is difficult to study in
humans, for both practical and ethical reasons. Therefore, neurobiologists interested
in neuronal plasticity tend to focus their research on monkeys, rodents, and, quite
frequently, invertebrates. Especially important for our understanding of synaptic
plasticity has been research on a marine snail called Aplysia californica (California
sea hare). Research on this species began in the late 1960s and was led by Eric Kandel,
who won the 2000 Nobel Prize in Physiology or Medicine.
Sensitization in Aplysia
Eric Kandel decided to study synaptic plasticity in Aplysia because this animal has a
relatively small number of very large neurons (see Box 3.1: The Impact of Invertebrates
on Neurobiology), which are much easier to study than neurons in the mammalian
hippocampus that Kandel had studied previously. Aplysia also exhibits various be-
haviors that can be modified as a result of experience. Most important for our pur-
poses is the gill withdrawal reflex. To understand this reflex, you need to know that
sea hares use a fleshy tube (siphon) to draw water over their gills. When an Aplysia is
relaxed, the siphon and gills are visible from above. However, when the siphon is
touched, both structures are withdrawn and covered by protective flaps. In nature,
this withdrawal reflex protects the delicate gills from rough seas or predators. In the
laboratory, the reflex can be triggered by a puff of water aimed at the siphon. Import-
antly, the gill withdrawal reflex can be triggered even when much of the body is dis-
sected away, leaving only siphon, gill, and tail as well as the neurons connecting those
body parts. Touching the siphon in such a semi-intact preparation causes the gill to
contract for a few seconds before it relaxes again. As shown in Figure 3.1, the neural
circuit underlying this behavior consists of sensory neurons that innervate the siphon
and synapse directly on large motor neurons, which innervate the gill muscles.
How Are Synapses Strengthened intheMarineSnail Aplysia? 73
Neural Mechanisms of Sensitization A
Sensitization of the gill withdrawal reflex occurs when an Touch stimulus
Aplysia receives a noxious (potentially harmful) stimulus
on its external body surface, most commonly the tail. After
such a stimulus, the gill withdrawal reflex is potentiated Noxious Siphon Sensory
(strengthened) in the sense that a light touch on the siphon stimulus Serotonergic
now causes the gill to be withdrawn for much longer than neuron neuron
before (Figure 3.1). The animal becomes “sensitized” to
future threats after experiencing the noxious stimulus. As
Kandel and his collaborators discovered, one neural cor- Recording
electrode
relate of this behavioral sensitization is a slight increase in Tail Sensory Motor
the duration (broadening) of the action potentials that neuron neuron
sensory neurons generate in response to a touch of the
siphon. Broadening the action potentials causes more calcium Gill
ions to flow into the presynaptic terminal, which increases
the amount of neurotransmitter (glutamate) that is re-
leased onto the motor neuron. Increased transmitter re-
lease increases the amplitude of the excitatory postsynaptic
potentials (EPSPs) in the motor neuron by more than
100% (Figure 3.1 C). Because the change in EPSP amplitude B Withdrawal reflex C Motor neuron EPSPs
results mainly from an increase in transmitter release, the 200
change is said to be presynaptic. Later in this chapter you Median duration (s) Post 1.5
Median amplitude (mV)
will see that changes in synaptic strength often involve 150 1.0
changes inside the postsynaptic cell, but sensitization of 100
the gill withdrawal reflex in Aplysia involves mainly pre-
synaptic alterations, notably action potential broadening
and increased transmitter release.
Central to the mechanisms underlying sensitization is 50 0.5
a set of neurons that are activated by noxious stimulation Pre Post Pre
of the skin and then release serotonin at several locations 0 Control Sensitized 0 Control Sensitized
in the Aplysia nervous system, including the presynaptic
terminals of the sensory neurons in the gill withdrawal
reflex circuit (Figure 3.1 A). Experiments have shown that
blocking serotonin receptors on those sensory neurons Figure 3.1 Sensitization of Aplysia’s withdrawal reflex.Depicted in
prevents the ability of noxious stimuli, such as tail shocks, (A)is a semi-intact Aplysia preparation. The bar graph in (B) reveals that
the mean duration of the gill withdrawal reflex increases after applying
to induce sensitization. Conversely, pumping a bit of sero- a painful stimulus to the animal’s tail. No such enhancement is seen in
tonin out of a micropipette onto the sensory neurons can control animals. Panel (C) shows that sensitization increases the ampli-
sensitize the gill withdrawal reflex, even if no tail shocks tude of motor neuron EPSPs elicited by touch of the siphon. [After
are applied. Together, these studies indicate that serotonin Kandel, 2001]
release is necessary and sufficient for sensitization of Aplysia’s gill withdrawal reflex.
In general, showing that a neural mechanism is necessary and sufficient for a particu-
lar behavior (or change in behavior) goes a long way to establishing a causal link
between the two.
The Role of Serotonin in Sensitization
How does serotonin change the sensory neurons? First, it binds to G protein-
coupled (see Figure 2.22) serotonin receptors on the sensory neuron. Activation
of these metabotropic receptors activates the enzyme adenylate cyclase, which
synthesizes cyclic adenosine monophosphate (cAMP) from ATP. Increased
cAMP levels within the sensory neuron then activate protein kinase A (PKA), an
enzyme that phosphorylates voltage-gated potassium channels in the neuronal
membrane (Figure 3.2). When these potassium channels are phosphorylated,
their probability of being open decreases. This reduces potassium influx and
therefore decreases the rate at which the cell membrane is repolarized during the
falling phase of the action potential (see Chapter 2, Figure 2.7). As a result, the
duration of the action potential increases and, as we just discussed, this broadening
74 Chapter 3 Neuronal Plasticity
EVOLUTION IN ACTION
Box 3.1 The Impact of Invertebrates on Neurobiology
In Chapter 2 you learned about Hodgkin and Huxley’s influen- habituation, and classical (Pavlovian) conditioning. Although
tial work on squid giant axons. This research could not have Aplysia’s behavior is less complex than that of vertebrates,
been performed on the much thinner axons of mammals, yet the molecular mechanisms of learning and memory in this
it revealed principles of neuronal signaling that apply across humble species have turned out to be broadly conserved.
all species with nervous systems. In this chapter, you learned Another invertebrate species that plays a major role in
about landmark studies on another mollusk, the sea hare neurobiology is the fruit fly Drosophila melanogaster. Its main
Aplysia californica (Figure b3.1). This species has only about attraction for research is that mutant flies are fairly easy to
20,000 neurons, many of which have very large cell bodies, which generate and analyze. Taking advantage of these features, re-
facilitates molecular and physiological analyses. Aplysia also searchers generated multiple strains of learning-impaired fly
exhibits multiple forms of learning, including sensitization, mutants and then identified which genes had been altered.
They found that many of the Drosophila genes involved
in learning and memory are hom*ologs of the genes that
A had been identified as being important for learning and
Head memory in Aplysia. This discovery of molecular conserva-
tion across two invertebrate species that are relatively
distant relatives suggested that the same mechanisms
might be at work also in vertebrates. To a large extent,
they are.
Lobsters and other crustaceans have also advanced
neuroscience research significantly. In these species,
small groups of neurons generate rhythmic outputs that
control various aspects of digestion. By recording and
manipulating the activity of these neurons, researchers
1 cm discovered some general principles of how neurons
generate rhythmic activity and, more generally, of neural
B network dynamics. They discovered, for example, that
a neuron’s physiological properties may be altered by
Giant serotonin and other neuromodulators, and that this mod-
cell body ulation can profoundly alter network dynamics, even
though the network’s anatomy remains unchanged.
Invertebrates sometimes don’t get enough credit
for their contributions to neuroscience, presumably
because human neurological disorders are difficult to
study in invertebrates and because invertebrate ner-
vous systems are organized quite differently from ver-
tebrate nervous systems. Still, many problems are
much easier to study in invertebrates, and some of the
500 µm differences between vertebrates and invertebrate ner-
vous systems may be smaller than they at first appear.
Moreover, some invertebrates are cognitively quite
complex. For example, honeybees can grasp simple
Figure b3.1 The large neurons of Aplysia.The photograph in (A) de- concepts, such as “same versus different”; wasps can
picts the sea hare Aplysia californica. (B) shows its abdominal ganglion, recognize each other by their facial markings; and oc-
dissected free of surrounding tissues, although several nerves and topuses can learn to solve puzzles by watching each
major axon tracts remain attached. Some of the cell bodies are gigantic. other. Thus, many invertebrates deserve more recogni-
[Images courtesy of John H. Byrne] tion than they get.
of the action potential increases calcium influx into the axon terminal and boosts
transmitter release. Voila! This is how serotonin release in response to a noxious
stimulus strengthens the gill withdrawal reflex. Some additional mechanisms
are known to be involved as well, but they need not concern us here. Instead, let
us consider variations in how long sensitization lasts.
How Are Synapses Strengthened intheMarineSnail Aplysia? 75
BRAIN EXERCISE Apply painful stimulus to skin
Can you think of an experimental manipulation that
would prevent short-term sensitization in Aplysia? Tail sensory neuron Serotonergic interneuron
Making Sensitization Last for Days Activate Release serotonin
sensory neuron from interneuron
The sensitization that we have discussed so far lasts a few
minutes or hours. However, if an Aplysia is given a series of
noxious stimuli, rather than just one, then the induced sen-
sitization is much more prolonged. For example, giving the Axon terminal of siphon sensory neuron
animal 5 tail shocks over a period of 4 days can enhance Phosphorylate Activate
the gill withdrawal for several weeks. In a semi-intact prep- potassium channels serotonin receptor
aration, a series of 5 serotonin pulses, spaced half an hour
apart, can trigger sensitization lasting more than 24 hours.
Incidentally, if the same number of shocks or serotonin
pulses is given over a much shorter interval, then the sensi- Broaden Activate
action potential adenylate cyclase
tization is less persistent. This phenomenon has a clear
parallel in your own life. Cramming for an exam the night
before is not nearly as effective as spacing those learning Increase Increase
sessions out over several days, at least if the goal is to re- calcium influx cyclic AMP level
member the information long term. This is a general psy-
chological principle: spaced learning is better than massed
learning at creating long-lasting memories. Increase Activate
glutamate release protein kinase A
The mechanisms underlying long-term sensitization
differ from those that cause short-term sensitization because
long-term sensitization requires protein synthesis, whereas
short-term sensitization does not. Applying a series of nox- Motor neuron
ious stimuli to a semi-intact Aplysia that is bathed in a drug
that inhibits protein synthesis fails to induce long-term Increase Increase motor
EPSP amplitude neuron firing rate
sensitization, even though short-term sensitization is intact.
Of course, short-term sensitization does involve proteins
(enzymes and ion channels), but the proteins used in short- Strengthen gill withdrawal reflex
term sensitization were synthesized before the sensitization
occurs. In contrast, long-term sensitization requires the trans-
lation of new proteins from mRNA. Figure 3.2 Mechanisms of short-term sensitization.Shown here is
the principal chain of events leading to short-term sensitization of
The Role of CREB in Long-term Sensitization Aplysia’s gill withdrawal reflex.
Intrigued by this discovery, researchers set out to determine which proteins must
be synthesized to generate long-term sensitization and what signals trigger their
synthesis. They found that repeated noxious stimulation leads to repeated release of
serotonin. As a result, the levels of activated PKA in the sensory neuron rise to such
a high level that some of the activated PKA makes its way into the sensory neuron’s
nucleus, where it phosphorylates cAMP response element binding protein (CREB).
This protein is a transcription factor, which means that it helps regulate the expres-
sion (transcription from DNA) of other genes. Specifically, phosphorylated CREB
binds to cAMP response element (CRE) sequences that are located upstream of
many genes. When bound to such a CRE sequence, the phosphorylated CREB inter-
acts with nearby RNA polymerases, which then begin their task of transcribing the
downstream gene (Figure 3.3). Soon thereafter, the transcribed mRNAs are trans-
lated into proteins and shipped from the nucleus to other parts of the neuron.
To show that CREB is a critical link in the induction of long-term sensitization,
experimenters injected synthetic oligonucleotides (short single-stranded RNA mol-
ecules) into an Aplysia sensory neuron and then applied several pulses of serotonin
to its synapse onto a motor neuron (Figure 3.3). When the injected oligonucleotides
contained the CRE sequence, and therefore bound any CREB molecules not yet
attached to DNA, then only short-term sensitization was induced. In contrast, when
76 Chapter 3 Neuronal Plasticity
A Serotonin-filled B Inject CRE
micropipette oligonucleotides
(5 spaced puffs) Recording
electrode
Sensory neuron
Motor neuron
Nucleus
Coding CREB-1 CRE
CRE Promoter region oligonucleotide
injection
CREB-1 Transcription
RNA polymerase mRNA
complex
Figure 3.3 Blocking long-term sensitiza- No oligonucleotides CRE oligonucleotides
tion.Repeatedly puffing serotonin onto 24 hrs after serotonin
the synapses between Aplysia sensory
andmotor neurons (A) increases EPSP Before serotonin
amplitude 24 hours later. When the same
experiment is performed after injecting Before serotonin 5 mV
CRE oligonucleotides into the cell body 5 mV
ofthe sensory neuron (B), long-term 24 hrs after serotonin
sensitization is blocked, presumably
because the oligonucleotides bind to the 25 ms 25 ms
CREB proteins and prevent them from
binding to the DNA. [After Kandel, 2001,
and Dash et al., 1990]
the injected oligonucleotides contained random sequences that do not bind CREB,
then both short-term and long-term sensitization were observed. Therefore, CREB is
necessary for the induction of long-term sensitization. In a complementary experi-
ment, researchers showed that an injection of phosphorylated CREB into the sens-
ory neuron is sufficient to trigger long-term sensitization. Overall, these studies show
that phosphorylated CREB is both necessary and sufficient for long-term sensitiza-
tion in Aplysia.
Many different genes have CRE sequences upstream of their promoters and are,
therefore, upregulated by phosphorylated CREB. Among the most interesting is a
gene that codes for ubiquitin hydrolase, which degrades the regulatory subunit of
PKA and thereby renders PKA persistently active. This enzymatic modification pro-
longs synaptic potentiation because it ensures that PKA remains activated long after
the serotonin release. Other genes induced by phosphorylated CREB are thought to
be involved in increasing the physical size of sensory axon terminals and the sprout-
ing of new synapses. Indeed, neuroanatomical studies have shown that the number of
synapses that each sensory neuron makes on a motor neuron more than doubles with
long-term sensitization.
Heterosynaptic versus hom*osynaptic Potentiation
Before we leave this discussion of sensitization in Aplysia, it is important to note that
the strengthened synapses in this form of learning are not the synapses that were ini-
tially activated. The noxious stimuli that trigger the sensitization activate the tail
sensory neurons and, through them, the serotonin neurons. They do not activate the
sensory neurons that carry information from the siphon to the motor neurons (see
Figure 3.1 A). Therefore, the synaptic potentiation responsible for sensitization in
Aplysia is heterosynaptic (involving different synapses), rather than hom*osynaptic
How Are Synapses Strengthened in Mammals? 77
(involving the same synapse). This distinction is worth making because Meynert,
James, and Cajal had envisioned synaptic potentiation to be hom*osynaptic. They
thought that only active synapses would be strengthened. As you will see, hom*osyn-
aptic potentiation is, indeed, quite common in mammalian brains. Nonetheless, it is
important to learn about sensitization in Aplysia because this work has revealed
molecular mechanisms, such as the role of CREB in regulating transcription of
learning-related genes, that are at work also in other species, including fruit flies and
mammals (see Box 3.1).
BRAIN EXERCISE
In your view, why might new protein synthesis be needed for long-term
sensitization but not for short-term sensitization?
3.2 How Are Synapses Strengthened in Mammals?
Research on the mammalian spinal cord in the 1950s revealed that high-frequency
repetitive stimulation of sensory spinal nerves can strengthen the responses of spinal
motor neurons to sensory inputs. One such experiment is shown schematically in
Figure 3.4. First, the experimenters record the EPSP that is generated by the motor
neuron in response to stimulating its sensory input once. Next, the experimenters
stimulate the input pathway at a rate of a hundred or so pulses per second. This kind
of high-frequency repetitive stimulation is called tetanic stimulation (derived from
the word tetanus, which refers to prolonged muscle spasms). During the initial phase
of the tetanic stimulation, the motor neuron’s membrane potential becomes increas-
ingly depolarized as a result of temporal summation (see Chapter 2). The membrane
potential then becomes slightly less depolarized (depressed) as the presynaptic neu-
rons run out of vesicles with neurotransmitter. At the end of the tetanic stimulation,
the motor neuron’s membrane potential gradually returns to its resting value. None
of this is very surprising. Half a minute later, however, the motor neuron’s response to
a single stimulus is much larger than it had been before the tetanic stimulation (post vs.
pre in Figure 3.4). This phenomenon is called post-tetanic potentiation, and it is
thought to result mainly from a buildup of calcium within the presynaptic terminal.
Although the discovery of post-tetanic potentiation was very
exciting, the phenomenon only lasts a few minutes and is,
therefore, insufficient to account for long-term memory. Pre Tetanic stimulation Post
Hippocampal Long-term Potentiation Temporal Depression
summation
In the 1960s, research on neuronal plasticity in mammals
shifted its focus from the spinal cord to the hippocampus Potentiation
(see Box 3.2) because by then, researchers knew that le- EPSP
sions of the hippocampus cause profound memory loss.
Two of the first neuroscientists looking for evidence of EPSP
memory-related synaptic plasticity in the hippocampus
were Timothy Bliss and Terje Lømo. In an influential ex- Pre Post
periment, they stimulated a set of axons in the perforant -100
path, which projects from the neocortex to a part of the 0 100 200 300 ms 30 sec
hippocampus termed dentate gyrus (Figure 3.5). Bliss
and Lømo also recorded the postsynaptic responses of the Figure 3.4 Effects of tetanic stimulation.In this idealized experiment,
dentate neurons to the perforant path stimulation. Their a series of electrical stimuli (blue vertical lines) is applied to an axon that
recordings reflected the sum of many different EPSPs but synapses onto a neuron, whose responses are recorded intracellularly
still indicated the strength of the synaptic connection (red traces). During repetitive high-frequency (tetanic) stimulation the
between the axons of the perforant path and the dentate membrane potential increases because of temporal summation and
neurons. Bliss and Lømo discovered that tetanic stimula- then decreases as the presynaptic cell runs low on releasable transmit-
tion of the perforant path in anesthetized rabbits increases ter. However, 30 seconds after the end of the tetanic stimulus, EPSP
the amplitude of the EPSPs in the dentate gyrus by more a mplitude is larger than it had been before the tetanic stimulation,
than 100% (Figure 3.6). Because this synaptic potentiation r evealing post-tetanic potentiation.
78 Chapter 3 Neuronal Plasticity
NEUROBIOLOGY IN DEPTH
Box 3.2 Hippocampal Structure and Functions
The hippocampus is one of the most intensively studied brain Lesions of the hippocampus in rats or mice impair their
areas. In rodents it is a relatively large, banana-shaped struc- ability to remember spatial locations relative to one another.
ture in the caudal half of the telencephalon, beneath the neo- Furthermore, many neurons in the rodent hippocampus are
cortex. In humans the hippocampus occupies a much smaller “place cells,” which increase their firing rate when the rat is in a
fraction of the telencephalon and looks somewhat like a particular location, regardless of how the animal is oriented.
seahorse (hippocampus means “sea-horse” in Ancient Greek). Therefore, we can conclude that a major function of the hippo-
Compared to a rodent’s hippocampus, the human hippocam- campus in rodents is spatial memory.
pus is shifted caudoventrally so that most of it lies within the
medial aspect of the temporal lobe. Because of this shift in Humans with hippocampus lesions also have trouble
p osition, the ventral hippocampus in rodents is topologically forming new spatial memories, but they have a much more
equivalent to the anterior hippocampus in humans, and a obvious general problem: they become unable to form new
rodent’s dorsal hippocampus corresponds to the human pos- episodic memories, defined as memories of what happened
terior hippocampus (Figure b3.2). when and where in a person’s own life. For example, when
a person with bilateral damage to the hippocampus meets
A Spinal cord
Dorsal hippocampus someone new, they soon forget that they had ever met
this person, even if they meet the same new person every
Ventral day for several weeks. Given these data, it seems as if the
hippocampus hippocampus has a larger set of functions in humans than
in rodents. In the latter, its function is spatial memory; in
Olfactory bulb the former, it has the far more general function of remem-
B bering what happened when and where.
Hippocampus The hypothesis that hippocampal functions differ sub-
stantially across species is reasonable, but some scientists
Temporal lobe suggest that the differences may be more apparent than
real. Because we cannot ask a rodent to tell us in words
Figure b3.2 The hippocampus in a rat and a human.In a rat’s brain, what it experienced yesterday, researchers must devise
here shown from an anterolateral perspective (A), the hippocampus clever experiments to test whether rodents might none-
(purple) lies in the caudal telencephalon. In a human brain, here theless remember what they did where and when. Such
shown from the same anterolateral perspective (B), the hippocampus experiments have shown that rodents can form some
lies mainly in the temporal lobe. types of episodic memory. Hippocampal lesions impair
those memories.
Another factor to consider is that scientists working
on rodents have studied mainly the dorsal hippocampus,
whereas studies on the human hippocampus focus mainly
on the anterior hippocampus. As noted earlier, the ante-
rior portion of the human hippocampus is topologically
equivalent to the rodent ventral hippocampus. Therefore,
it is possible that the suspected species differences be-
tween rodent and human hippocampi are not true spe-
cies differences but rather a consequence of functional
differences between different parts of the hippocampus
(dorsal vs. ventral and anterior vs. posterior). This hypoth-
esis is consistent with anatomical data showing that the
posterior hippocampus in primates receives more refined
spatial information than the anterior hippocampus and
with physiological data indicating that in rodents the
dorsal hippocampus encodes spatial information more
precisely than the ventral hippocampus. Unfortunately,
the functions of the ventral hippocampus in rodents and
the posterior hippocampus in humans remain quite poorly
understood. Therefore, the question of how the human
hippocampus differs from that of rodents is still a subject of
considerable debate.
How Are Synapses Strengthened in Mammals? 79
lasts for many hours, it came to be known as long-term Perforant
potentiation (LTP). Some studies have reported that hippo- path
campal LTP can last for several weeks or even months.
CA fields
To understand the mechanisms underlying LTP, research-
ers employed the brain slice technique, which involves dis- Dentate gyrus
secting the brain out of a deeply anesthetized animal, slicing Hippocampus
it into relatively thick slabs (~0.4 mm thick), and then bathing
each slice in a highly oxygenated solution containing nutri-
ents. Under these conditions, neurons can be kept alive for
many hours. Stimulating and recording electrodes can be
placed more precisely in brain slices than in intact brains,
intracellular recordings are simpler to obtain, and drugs that
block specific receptors or ion channels can be added to the
solution in which a slice is bathed.
Using the brain slice technique, experimenters discovered
that LTP can be elicited not only in the synapses of the per-
forant path, but also in several other hippocampal pathways.
A second major feature of hippocampal LTP is that it tends
to be input specific, meaning that only the stimulated, active
synapses are strengthened. Therefore, in contrast to sensiti-
zation in Aplysia, hippocampal LTP is hom*osynaptic. An-
other difference between Aplysia and mammals is that LTP in
the mammalian hippocampus involves mainly postsynaptic
changes, whereas sensitization in Aplysia depends primarily
on presynaptic modifications.
Hebbian Long-term Potentiation Figure 3.5 Section through a rat hippocampus.Shown at the top is
So far we have discussed LTP as if strong, repetitive firing a coronal section through a rat brain. The bottom diagram shows
of a neuron strengthens all the synapses that this neuron some major hippocampal divisions and connections. The yellow
makes onto other neurons. This is not what Meynert and neuron projects from the dentate gyrus to one of the cornu ammonis
James had in mind when they proposed their model of how
(CA) fields. The orange neuron projects to the dentate gyrus through
a child learns not to touch a flame. They had proposed that the perforant path. [Brain section image from brainmaps.org]
strengthening occurs only in the connections between neu-
rons whose activity is associated (linked) in time. This idea was formalized by Donald
Hebb in 1949. As part of an ambitious effort to understand the neural basis of thought
and memory, Hebb proposed the following:
When an axon of cell A is near enough to excite a cell B and repeatedly or persistently
takes part in firing it, some growth process or metabolic change takes place in one or
both cells such that A’s efficiency, as one of the cells firing B, is increased.
(Hebb, 1949, p. 62).
EPSP amplitude 4 tetanic stimulus trains Long term potentiation (LTP)
(% of pre-stimulation) 300
Figure 3.6 Long-term potentiation
200 (LTP).Bliss and Lømo (1973) recorded syn-
aptic responses in the dentate gyrus of a
100 Pre-stimulation amplitude rabbit’s hippocampus while applying four
high-frequency (tetanic) trains of electrical
0 stimuli to the perforant path (see Fig. 3.5).
0 2 4 6 8 10 12 14 Already after the first tetanic stimulus,
Time (hours) EPSP amplitude increased almost 100%.
Importantly, some synaptic potentiation
persisted for more than 10 hours after the
stimulation. [After Bliss and Lømo, 1973]
80 Chapter 3 Neuronal Plasticity
A Inject current into The key idea in this statement, commonly referred to
postsynaptic cell as Hebb’s rule, is that synapses should be strengthened
Stimulate only if they were active when the postsynaptic neuron was
presynaptic axon depolarized enough to fire an action potential. If a synapse
is active but the postsynaptic cell does not fire, then this
EPSP size (%)B synapse should not be strengthened. Conversely, even weak
synapses should be potentiated if they were active just
200 Postsynaptic depolarization before the postsynaptic neuron fires an action potential.
A simplified version of Hebb’s rule states: Neurons that fire
100 together, wire together. Hebb had no direct evidence for such
Postsynaptic hyperpolarization plasticity, but we now know that LTP at many synapses
does follow Hebb’s rule.
-5 Tetanic 5 10 15 20
stimulus Time (min) To test whether LTP at a specific set of synapses follows
Hebb’s rule, neuroscientists can stimulate presynaptic
Figure 3.7 Testing whether LTP obeys Hebb’s rule.In this idealized axons while depolarizing or hyperpolarizing the post-
experiment, one electrode is used to stimulate presynaptic axons synaptic cell by means of intracellular current injections
repeatedly (tetanically). A second electrode is used to inject either (Figure 3.7). Depolarizing the postsynaptic cell during the
depolarizing or hyperpolarizing current into the postsynaptic cell presynaptic stimulation mimics the situation in which the
duringthe presynaptic stimulation. Every 30 seconds the postsynaptic presynaptic cell “takes part in firing” the postsynaptic cell,
neuron’s response to a single presynaptic stimulus is recorded. If the whereas hyperpolarizing current injections mimic the
EPSPs increase in amplitude only when the stimulation is coupled with situation in which the presynaptic stimulation does not
d epolarizing current injections, and not when it is coupled with hyper- participate in firing the postsynaptic cell. Hebb’s rule
polarization, then the synaptic potentiation obeys Hebb’s rule. predicts that LTP should be observed only when the
presynaptic stimulation is accompanied by postsynaptic
depolarization. Using such experiments, researchers have
shown that LTP at many synapses does, indeed, obey
Hebb’s rule. This form of LTP is sometimes termed
Hebbian LTP or associative LTP; but we can simply call it
LTP, keeping in mind that not all instances of LTP are
necessarily Hebbian.
BRAIN EXERCISE
In the statement “neurons that fire together, wire
together,” what exactly does “wire together” mean?
Mechanisms of LTP Induction
As mentioned earlier, short-term sensitization in Aplysia involves changes in existing
proteins, whereas long-term sensitization requires new protein synthesis. A similar
distinction applies to mechanisms underlying LTP in mammals, except that neuro-
scientists don’t talk about short-term and long-term LTP (these terms would be con-
fusing). Instead, neurobiologists distinguish between the induction (triggering) of
LTP and its stabilization.
Role of the NMDA Receptor
A central player in the molecular cascade that induces LTP is the NMDA receptor
(see Chapter 2). In contrast to the AMPA-type glutamate receptor, NMDA-type glu-
tamate receptors can open in response to glutamate only when the postsynaptic cell
is depolarized because magnesium ions block the receptor’s central pore as long as
the cell membrane is near its resting potential (Figure 3.8). As the postsynaptic cell
becomes depolarized, the magnesium block is removed and the NMDA channel
can open in response to glutamate, letting both calcium and sodium ions flow into
the postsynaptic cell. Thus, you can think of the NMDA receptor as a molecular
mechanism for detecting the coincidence of presynaptic glutamate release and post-
synaptic depolarization. This molecular coincidence detection allows a cell to deter-
mine whether an active synapse has taken part in firing the postsynaptic cell. It
would, therefore, be a good mechanism for triggering LTP.
How Are Synapses Strengthened in Mammals? 81
A Near the resting potential B Postsynaptic depolarization
Presynaptic
axon Figure 3.8 The NMDA receptor as a
m olecular trigger for LTP.If the post
Glutamate synaptic membrane is near its resting
potential (A), then NMDA receptors are
Mg2+ ion blocked by magnesium (Mg21) ions.
Theseions are dislodged when the post-
AMPA Mg2+ ion Na+ synaptic cell is strongly depolarized (B).
receptor Na+ NMDA Ca2+ Once the magnesium block is gone,
Additional Na1and Ca21 ions can flow through the
receptor AMPA receptors Calmodulin NMDA receptor when glutamate is bound.
Postsynaptic cell An increase in postsynaptic calcium then
CaMKII triggers an intracellular signaling cascade
that ultimately leads to the insertion of
additional AMPA receptors into the post-
synaptic membrane, which strengthens
the synapse.
A potential problem with using NMDA receptors to trigger LTP is that a post
synaptic cell expressing only NMDA receptors would be unable to open in response
to glutamate. As noted earlier, the NMDA receptors cannot open until the cell be-
comes depolarized; but as long as the NMDA receptors remain closed, the depolar-
ization cannot get going. Such “silent synapses” have been observed in circuits that
are still developing, but they would not work well in adult nervous systems. Most
neurons solve this silent synapse problem by populating their postsynaptic mem-
branes with AMPA as well as NMDA receptors. Because AMPA receptors have no
magnesium block, they can open in response to glutamate even near the cell’s resting
potential. Thus, they allow the postsynaptic depolarization to get off the ground. Once
the postsynaptic cell has become sufficiently depolarized, the magnesium block is
removed from the NMDA receptors, allowing them to open as well.
How does the opening of postsynaptic NMDA receptors cause changes in
synapse strength? Extensive studies, mainly on hippocampal slices, have shown that
calcium influx through the open NMDA receptors is a critical factor. When postsyn-
aptic calcium levels rise, a protein called calmodulin binds calcium, and the resulting
calcium/calmodulin complex activates a protein kinase termed calcium/calmodulin
kinase II (CaMKII). Activated CaMKII promotes the insertion of additional AMPA
receptors into the postsynaptic membrane (Figure 3.8 B). It also phosphorylates
the AMPA receptors, which increases the rate at which ions can flow through them
(the probability of the channels being open is reportedly unchanged). The upshot
of these modifications is that the postsynaptic cell becomes more responsive to pre-
synaptic glutamate release. For a given amount of released glutamate, the cell’s EPSP
increases in amplitude. Importantly, the synaptic potentiation is accomplished by
changes in the postsynaptic cell, not by increased presynaptic transmitter release
(which, as you may recall, is the basis of synaptic potentiation in the gill withdrawal
reflex of Aplysia). There is some evidence for presynaptic changes also in LTP, but
these effects are less well understood than the postsynaptic modifications.
The Role of Synapse Growth
The intracellular processes that induce LTP are the kind of “metabolic change” that
Hebb proposed as one possible mechanism for strengthening synapses. The other
mechanism Hebb envisioned is synapse growth. Indeed, an interesting study by
Masanori Matsuzaki and his collaborators has shown that synapses may grow quite
rapidly during LTP induction (Figure 3.9). To understand this somewhat compli-
cated experiment, you need to know that the dendrites of many neurons are covered
with tiny spines (see Figure 2.20). Such “spiny neurons” receive most of their excit-
atory input onto the tips of their spines, allowing you to think of each spine as the
postsynaptic side of a synapse. Taking advantage of this arrangement, Matsuzaki
82 Chapter 3 Neuronal Plasticity
A Use laser to “uncage” glutamate locally could determine whether synapses grow larger when they
are strengthened by measuring changes in the size of den-
small spine dritic spines after LTP induction. To induce LTP at a spe-
cific spine, the researchers labeled a postsynaptic neuron so
- 3 min 2 min 27 min 1 µm that the spines could be visualized under the microscope.
81 min Next, they stimulated the glutamate receptors on a specific
B spine by bathing the tissue in caged glutamate (glutamate
500 Glutamate uncaged molecules inside a molecular “cage”) and using a laser to
“uncage” some of the glutamate next to one spine. The
300 Spine volume (%) Transient uncaged glutamate diffuses across the synaptic cleft and
binds to postsynaptic glutamate receptors. To ensure that
100 Volume increase (%) Persistent the glutamate could open the postsynaptic NMDA re-
-40 ceptors, the experimenters bathed the tissue (a hippo-
0 40 80 campal slice) in a magnesium-free solution, which prevents
C Time (min) magnesium block.
200
100 Transient The results of Matsuzaki’s experiment are shown in
Persistent Figure 3.9. Within 1–5 minutes after the glutamate re-
lease, the volume of the stimulated spine doubles, on aver-
age. Spine volume then decreases, but the stimulated
spines remain enlarged for more than an hour. Impor-
tantly, bathing the tissue in drugs that block NMDA
receptors or calmodulin prevents spine growth. Therefore,
spine growth seems to be triggered by the same molecular
processes that are needed for LTP, suggesting that spine
growth contributes causally to LTP induction.
BRAIN EXERCISE
Why is it important that Matsuzaki et al. in their spine
growth experiment bathed the hippocampal slice in a
magnesium-free solution?
0 Block Block Block Mechanisms of LTP Stabilization
Control NMDA calmodulin CaMKII
As you learned in the previous section, LTP induction
Figure 3.9 LTP and spine growth.A laser was used to release caged involves the activation of CaMKII, which then boosts
glutamate at one spine of a fluorescently labeled hippocampal neuron AMPA receptor number and ion f low in the activated
(A). The graph in (B) shows that spine volume increased dramatically synapses. However, CaMKII stays active for only about
right after the glutamate release. Although some of the volume increase a minute, and the AMPA receptors that were inserted
was transient, some of it persisted for at least 100 minutes. The tissue into the postsynaptic membrane are soon removed again.
was bathed in a magnesium-free solution to boost the ability of gluta- Even spine growth is largely transient, especially when
mate to open NMDA channels. Panel (C) shows that spine growth does CaMKII is blocked (Figure 3.9 C). Yet studies with
not occur when NMDA receptors or calmodulin are blocked pharmaco- chronically implanted electrodes have shown that LTP can
logically. Blocking CaMKII prevents persistent enlargement without sometimes persist for months. We must wonder, therefore,
blocking transient growth. [After Matsuzaki et al., 2004] how changes in synaptic strength are stabilized for the
long term.
The Role of Protein Synthesis
The stabilization of LTP requires protein synthesis because protein synthesis inhibi-
tors prevent LTP from developing, allowing only for transient synaptic strengthen-
ing. This is not surprising, given the need for protein synthesis in Aplysia’s long-term
sensitization, but ask yourself, how can proteins that are synthesized in the nucleus
of the postsynaptic cell be used to strengthen only those synapses that contributed
to firing the postsynaptic cell, as Hebb’s rule requires? Protein synthesis usually
occurs in a cell’s nucleus, and the newly synthesized proteins are then shipped from
the nucleus into the rest of the cell, including the dendrites. How do those proteins
know which synapses they ought to fortify and which they ought to leave untouched?
One answer to this question is that synapses that were active during a postsynaptic
How Are Synapses Strengthened in Mammals? 83
Ribosomes Ribosomes
Spine Synapse Figure 3.10 The machinery for protein
Dendrite synthesis in dendrites.These transmission
electron micrographs show several small
Ribosomes clusters of ribosomes in the dendrites
500 nm (shaded light blue) of rat hippocampal
neurons. Because ribosomes are used to
generate proteins from mRNA, these data
strongly suggest that proteins can be
synthesized not only in the cell body but
also in dendrites, close to the synapses.
[From Steward and Levy, 1982]
depolarization can be marked with some sort of molecular “synaptic tag.” Newly syn-
thesized proteins are thought to recognize those tags and go to work selectively at
those tagged synapses. Although there is some evidence for this synaptic tagging
hypothesis, its details remain unclear.
Dendritic Protein Synthesis A Stimulate
Another mechanism for deploying newly synthesized Record from axons
proteins selectively at synapses that were active during isolated dendrites
a postsynaptic depolarization is to synthesize the needed
proteins out in the dendrites, right next to the synapses Cell body
that are to be strengthened. Early evidence for dendritic Knife cut
protein synthesis came from the anatomical observation
that a few ribosomes, which translate mRNA into protein, Dendrites Axon
are located inside the dendrites of hippocampal neurons
(Figure 3.10). These ribosomes are typically found at the EPSP (% of baseline)B Control
base of dendritic spines or sometimes inside a spine. To 200
demonstrate that these dendritic ribosomes synthesize
proteins that are needed to stabilize LTP, researchers used 150
a knife to separate hippocampal dendrites from their cell
bodies (Figure 3.11). In such isolated dendrites, LTP can Protein synthesis blocked
be induced by tetanic stimulation of the input pathway. 100
However, if the dendrites are bathed in a protein synthesis
inhibitor, then only transient synaptic potentiation is seen Stimulate to induce LTP
(Figure 3.11 B). These findings imply that LTP induction
causes ribosomes near the activated synapses to synthesize 0 20 40 60 80
new proteins, which are then used to stabilize the induced Time (minutes)
changes in synaptic strength.
Figure 3.11 Dendritic protein synthesis is needed for LTP.The den-
Although dendritic synthesis is known to play a role drites of hippocampal neurons were separated from their cell bodies by
in stabilizing LTP, which proteins are involved remains a knife cut (A). Axons terminating on those dendrites were stimulated
unclear. One likely candidate is CaMKII, whose role in repeatedly, leading to long-term potentiation (LTP) in the stimulated
adding AMPA receptors to the postsynaptic membrane we pathway, as demonstrated by a persistent increase in EPSP amplitude
already discussed. Another candidate is activity-related (B). Blocking protein synthesis (with emetine) permits transient poten-
cytoskeletal protein (ARC). Strong synaptic activity tiation (red) but prevents LTP stabilization. [After Cracco et al., 2005]
rapidly increases Arc gene expression in the cell nucleus.
Much of the newly transcribed Arc mRNA is then
shipped from the nucleus into the dendrites, where it
is translated. Because blocking Arc translation impairs
LTP and some forms of long-term memory, we can con-
clude that ARC is probably involved in stabilizing LTP.
EPSP (% Control pathway
100
84 Chapter 3 Neuronal Plasticity Stimulate to induce LTP
024 6 8
Time (hours)
A Potentiated pathway Zeta inhibitory B Control peptide
200 peptide (ZIP) 250
EPSP (% of baseline) EPSP (% of baseline) Potentiated pathway
200
150 150
Control pathway Control pathway
100 100
Stimulate to induce LTP Stimulate to induce LTP 6 8
024
024 68
Time (hours) Time (hours)
B
Figur2e530.12 Reversing late-stage LTP with zeta inhibCitoonrtyropl peeppttiidde (ZIP).Five hours after LTP induction, hippocampal neurons were bathed inEPSP (% of baseline)
ZIP(A). This treatment rapidlPyoetelinmtiainteadtepdatthhweaLyTP, whereas application of a scrambled (control) version of the same peptide had no effect on LTP (B).
EPSPs elicited by test stimuli applied to a second (control) pathway that had not been stimulated tetanically was not potentiated and not affected
by eit2h0er0peptide. [After Serrano et al., 2005]
150 Most likely, however, LTP stabilization involves a wide variety of proteins besides
Control pathway CaMKII and ARC.
100 Other Mechanisms Involved in LTP
Recent studies have shown that LTP can be destabilized by zeta inhibitory peptide
Stimulate to induce LTP (ZIP). When this peptide is applied to potentiated synapses as much as 5 hours after
0 2 (ho4urLsst)TrePngi nthd6 u(Fc tigiounre, s8ynaptic strengths rapidly return to their prestimulation levels of
Time 3.12). In essence, ZIP destroys the synaptic “memory.” No one is sure
how ZIP accomplishes this feat. Some data indicate that ZIP interferes with protein
kinase M-zeta (PKM-zeta), an enzyme that enhances synaptic transmission through
AMPA receptors. However, mice engineered to lack PKM-zeta still exhibit LTP and
are not impaired in several forms of learning and memory. Therefore, ZIP must also
interfere with other mechanisms that are crucial for LTP stabilization. The identity
of those other mechanisms is presently unknown.
Finally, the stabilization of LTP involves epigenetic changes, defined as the
chemical modification of DNA and its associated proteins, which lead to long-lasting
changes in gene expression. For example, drugs that prevent the de-acetylation of
histones, around which DNA strands are wound, can convert transient synaptic en-
hancement into LTP and make some memories more persistent.
BRAIN EXERCISE
Are the “synaptic tagging” and “dendritic protein synthesis” hypotheses mutually
exclusive? Why or why not?
3.3 When Are Synapses Weakened?
Although most of the research on synaptic plasticity has focused on the strengthen-
ing of synapses, brains must also have mechanisms for making synapses weaker.
Otherwise, if synapses could only get stronger, neural activity would soon spread
wildly through the brain, causing indiscriminate and excessive action potential
firing. Indeed, neurobiologists have found numerous examples of synapses being
weakened (depressed) for many hours after certain patterns of presynaptic stimula-
tion. This phenomenon is called long-term depression (LTD).
Cerebellar Long-term Depression
A particularly interesting form of LTD does the opposite of what Hebb had predicted:
it causes cells that “fire together” to become less strongly interconnected. Such
Climbing Purkinje cell
fiber When Are Synapses Weakened?cG erlalsnule
85
A Cerebellar circuitry
B Cerebellar LTD
Parallel fibers 1.5
Synaptic strength Stimulate
Synaptic strength PF and CF
1.0
Purkinje cell 0.5
Climbing 0 20 40 60
fiber Time (min)
Granule
cells
B Cerebellar LTD
Figure 3.13 Long-term depression (LTD) in the cerebellum.Panel (A) depicts the climbing and parallel fiber inputs to a cerebellar Purkinje cell.
dG eracpreha1(sB.e5)isnhtohwessStthrteiamntguretlahpteoetfittihvee coincident stimulation of the parallel and climbing fiber (PF and CF, respectively) inputs leads to a persistent
stimulated parallel fiber-Purkinje cell synapses. This synaptic weakening is called long-term depression. [After
Finchet al., 2012P] F and CF
1.0
anti-Hebbian plasticity is found in the cerebellum, whose functions we will discuss
extensively in Chapter 10. Briefly, the cerebellum fine-tunes movements (and, more
contr0o.v5ersially, some cognitive processes) by learning from errors and then adjust-
ing its output accordingly.
As shown in Figure 3.13 A, the cerebellum contains some very large neurons,
called Purkinje cells0, and a mu2c0h larger n40umber of6v0ery small neurons, called cerebel-
lar granule cells. Each PurkinTjiemceel(lmreince) ives input from thousands of granule cells,
whose axons are known as parallel fibers. In addition, each Purkinje cell receives
input from a climbing fiber, which wraps itself around the Purkinje cell dendrites.
When a climbing fiber fires an action potential and releases glutamate, the postsyn-
aptic Purkinje cell becomes strongly depolarized and fires an action potential of its
own. In contrast, the synapse between parallel fibers and Purkinje cells is relatively
weak. Glutamate release at these synapses tends to evoke small postsynaptic depo-
larizations that trigger action potentials only when many parallel fibers are activated
simultaneously. Most important for our present discussion is that activating the par-
allel and climbing fiber inputs simultaneously, and doing so repeatedly (~1/sec for
several minutes), causes the parallel fiber input to decrease in strength (Fig. 3.13 B).
This anti-Hebbian plasticity develops gradually and can last for many hours.
The mechanisms underlying cerebellar LTD resemble those of LTP but differ in
some important respects. As in LTP, calcium influx into the postsynaptic cell plays a
crucial role in cerebellar LTD. Specifically, the postsynaptic depolarization triggered
by climbing fiber activation causes calcium influx through voltage-gated calcium
channels and NMDA receptors at the climbing fiber synapses. Parallel fiber synapses
do not contain NMDA receptors, but they do contain metabotropic glutamate re-
ceptors (see Chapter 2, Fig. 2.22). When these metabotropic receptors are activated
by repetitive stimulation, they trigger an intracellular cascade (involving inositol
trisphosphate, abbreviated IP3) that ultimately leads to the release of calcium ions
from intracellular organelles. Although both climbing fiber and parallel fiber activa-
tion lead to increased calcium levels within the postsynaptic cell, coincident activation
of both the parallel and the climbing fibers increases calcium levels much more than
either form of stimulation can accomplish by itself. The sharp rise in postsynaptic
calcium then triggers persistent activation of an enzyme called protein kinase C,
which is involved in the removal of AMPA receptors from the postsynaptic side of the
synapse. As you can imagine, the removal of AMPA receptors weakens the synapse.
86 Chapter 3 Neuronal Plasticity
Change in EPSP amplitude (%) Postsyn. depol. Postsyn. depol. Spike Timing-dependent Plasticity
before after
Although LTD has been studied most thoroughly in the
presyn. stim. presyn. stim. cerebellum, LTD also occurs in several other brain re-
gions, including the hippocampus. Most interesting are
80 synapses that exhibit LTP if postsynaptic depolarization
follows presynaptic activity and LTD if postsynaptic de-
40 Potentiation polarization precedes presynaptic activity (Figure 3.14).
Such spike timing-dependent plasticity has been ob-
0 served in several different pathways, including some in
the hippocampus, but it is not ubiquitous. In general, the
-40 Depression type of plasticity that synapses can exhibit varies across
-80 -40 0 40 80 cell types. This variability results from many factors,
Time difference (ms) including differences in the transmitter receptors, ion
channels, and intracellular enzymes. As we discussed in
the previous chapter, neurons vary in many different re-
spects. One of these is how their connections change in
response to activity.
Figure 3.14 Spike timing-dependent plasticity.Using an experimen- BRAIN EXERCISE
tal design similar to that shown in Figure 3.7, scientists discovered that
whether the synapses between two cultured hippocampal neurons are What does it mean for synaptic plasticity to be
strengthened (potentiation) or weakened (depression) depends on the “anti-Hebbian”?
relative timing of presynaptic stimulation and postsynaptic depolariza-
tion. [After Bi and Poo, 1998]
3.4 Can Inactive Neurons Strengthen Their Inputs?
So far we have discussed changes in synaptic strength that are linked to specific pat-
terns of neural activity, but less specific forms of synaptic plasticity also occur. For
example, neurons that have been inactive for some time tend to increase the strength
of their excitatory synaptic inputs. In contrast, neurons that
A Control
have been firing at a high rate tend to become less respon-
sive. Strong evidence for such synaptic scaling comes from
studies with cortical neurons that were maintained in cell
Control = 100% culture for several days. Neurons cultured for 48 hours in
Individual EPSC Average EPSC a drug that blocks voltage-gated sodium channels (tetrodo-
toxin) exhibit larger excitatory postsynaptic currents
(EPSCs), which are analogous to EPSPs except that they mea-
B 48 hours in sodium channel blocker (TTX) sure current rather than membrane voltage, than neurons in
control cultures without tetrodotoxin (Figure 3.15). Con-
versely, neurons cultured in a solution containing bicuc-
culine, which blocks GABA receptors and therefore
increases neuronal firing rates, exhibit reduced EPSCs
192% of control
(Figure 3.15 C). The mechanisms underlying these activity-
dependent increases and decreases in synaptic strength
are poorly understood but probably involve the addition or
C 48 hours in GABA receptor blocker (bicucculine) removal, respectively, of postsynaptic AMPA receptors.
Why do neurons scale up their sensitivity to synaptic
input after they have been inactive and, conversely, scale it
10 pA 70% of control down after firing frequently? Because it would be wasteful
1 sec 10 ms to have neurons in your brain that are capable of generat-
ing action potentials but never receive enough excitatory
input to reach the action potential threshold. Synaptic
Figure 3.15 Synaptic scaling.Excitatory postsynaptic currents scaling ensures that such neurons do not remain silent for
(EPSCs) were recorded from voltage-clamped cortical neurons main- long. Conversely, it is dangerous for neurons to be overly
tained in cell culture (downward deflections indicate depolarizing
current). After 48 hours of exposure to tetrodotoxin (TTX), which elimi- sensitive to excitatory synaptic input and, therefore, firing
nates all action potentials, average EPSC amplitude was increased too frequently. Such hypersensitivity can trigger epileptic
dramatically. In contrast, after blocking GABA receptors for 48 hours, seizures through runaway excitation. It can also kill neurons
EPSC amplitudes decreased. Thus, the neurons scale their responses to through excitotoxicity (see Chapter 5). Synaptic scaling
glutamate release up or down, depending on how much they have mitigates these risks.
been firing. [From Turrigiano et al., 1998]
Can Experiences Rewire the Brain? 87
BRAIN EXERCISE
In what sense is synaptic scaling “anti-Hebbian”?
3.5 Can Experiences Rewire the Brain?
Neuronal plasticity may involve changes in the strength of existing synapses, but it
may also involve the formation of entirely new dendritic spines, axon terminals, and
major axonal branches. Such neuronal “rewiring” frequently occurs in the aftermath
of brain damage, but it also occurs in the context of learning and memory.
Turnover of Dendritic Spines
Using fluorescent markers to label spiny neurons, researchers have found that
dendritic spines are surprisingly dynamic. New spines are formed quite frequently
and older spines are lost. This high rate of spine turnover implies that the synapses
between neurons are more dynamic than people once believed. Most importantly,
changes in the rate of spine turnover may be linked to learning. For example, spiny neu-
rons that control singing in songbirds reduce their rate of spine turnover (Figure 3.16)
when young birds learn to sing. Thus, the capacity for learning seems to be associated
with a high degree of spine plasticity, whereas the act of learning involves the stabili-
zation of newly formed spines.
Sprouting of Axonal Connections
If each spine represents the postsynaptic side of a synapse, then the existence of spine
turnover implies that the presynaptic elements, the axon terminals, are also more
dynamic than once thought. It is technically difficult to document small changes in
axon terminals, but some studies have shown that axons can indeed sprout new
branches, even in intact, adult brains.
A good example of axon sprouting comes from mice undergoing eye-blink condi-
tioning. In this learning paradigm, animals are repeatedly presented with a tone,
at the end of which a noxious puff of air is delivered to one of the eyes. The animals
soon learn to associate the air puff with the tone and learn to blink when the tone is
presented alone. This eye-blink conditioning involves the sprouting of new axon
branches that terminate in the deep cerebellar nuclei, which lie beneath the cerebellar
cortex (where the Purkinje cells are located).
To reach this conclusion, experimenters injected an axon tracer into a hindbrain
region called the pons, which contains numerous auditory neurons. In untrained or
pseudoconditioned animals (mice presented with unpaired air puffs and tones), the
0 hours 0 hours 24 hours
20 µm 2 hours 0 hours 2 hours Figure 3.16 Dendritic spine turnover.
0 hours Shown in the top row is a spiny neuron in
New the brain of an adult zebra finch, shown
Lost Lost at different magnifications and different
2 µm times. The left 2 images in the bottom
row are close-ups of one dendrite, imaged
2 hours apart. You can see that one small
spine was lost during this interval. The
other images in the bottom row are from
another neuron, imaged similarly. In this
case, one spine disappeared while another
one sprouted anew. As young birds learn
to sing, the rate of spine turnover de-
creases (at least for neurons that had a
high turnover rate to begin with). [From
Roberts et al., 2010]
88 Chapter 3 Neuronal Plasticity
A Auditory projections from pons to cerebellar nuclei
Figure 3.17 Axon sprouting during Untrained Pseudoconditioned Conditioned
eye-blink conditioning.Axon tracers
were used to visualize axons that project 1 mm
from the auditory region of the pons to
the deep cerebellar nuclei (CN). Panel (A) Left CN Right CN Left CN Right CN Left CN Right CN
shows that these projections are more
extensive in eye-blink conditioned mice B Axon varicosities C
than in pseudoconditioned or untrained 1.0
mice. Panel (B) depicts some labeled axons Varicosities / labeled axon Left Right
with varicosities, which are most likely 3
synapses. The graphs in (C) show that the
number of varicosities in the cerebellar 10 µm 0.5 2
nuclei is significantly increased in the 0 1
c onditioned animals, relative to the con- Un Ps Cl 0
trols. To control for the size of the tracer
injection, the number of varicosities was Un Ps Cl
divided by the number of labeled axons in
the main fiber tract connecting the pons
to the cerebellum. [From Boele et al., 2013]
pons projects only weakly to the deep cerebellar nuclei. In contrast, after eye-blink
conditioning, the projections from the pons to the deep cerebellar nuclei are much
more extensive (Figure 3.17 A). They also exhibit more varicosities, which are most
likely sites of neurotransmitter release (Figure 3.17 B). Overall, these data strongly
suggest that eye-blink conditioning causes new axon branches to sprout and form
new synapses. It is not yet clear, however, whether these learning-related changes are
necessary or sufficient for eye-blink conditioning. They may be a correlate or conse-
quence, but not a cause.
BRAIN EXERCISE
Go back to the quote from Ramón y Cajal at the beginning of this chapter. How
well has his hypothesis been supported by modern data? What kind of plasticity
did Cajal overlook in his statement?
Sensory Cortex Plasticity
In most cortical areas, a given stimulus tends to activate a spatially coherent cluster
of neurons rather than a scattered set of isolated cells. This organizational feature
allows neuroscientists to construct “maps” of how sensory stimuli are represented in
the cortex by recording the stimulus preferences of neurons at several systematically
varied locations across the cortex. One such map is shown in Figure 3.18. It shows
that tones of a specific frequency tend to activate neurons in a specific part of the
primary auditory cortex. Moreover, the sound frequency to which cortical neurons
respond most strongly varies systematically across the auditory cortex, creating what
is called a tonotopic map. This map turns out to be plastic. Strong evidence for this
plasticity is that the tonotopic maps in naïve, untrained rats looks very different from
those in thirsty rats that were trained to associate a specific tone with the availability
of water (Figure 3.18). In such trained rats, the frequency of the conditioned stimulus
(the tone) is overrepresented relative to naïve rats, while other frequencies are
underrepresented. These learning-related changes in the tonotopic map are driven by
changes in how individual neurons respond to auditory stimuli. In the case illus-
trated in Figure 3.18, neurons that responded most robustly to frequencies other
than 6 kHz before training gradually shift their response preferences toward 6 kHz
as the animals learn the association between the 6 kHz tone and water availability.
The greater the number of cortical neurons that shift their response preference
toward the conditioned stimulus, the more this stimulus becomes overrepresented
Can Experiences Rewire the Brain? 89
A Tonotopic map in auditory cortex
Naïve rat Best Trained rat
frequency Trained frequency = 6 kHz
50 kHz
40
30
20
10
1 mm 0
B Training alters the tonotopic map Figure 3.18 Learning-related plasticity
in the auditory cortex.Thirsty rats were
50 Naïve 50 Trained Conditioned trained to press a bar for water whenever
rat rat frequency they heard a 6 kHz tone. The experiment-
= 6 kHz ers then compared the stimulus prefer-
40 40 ences of neurons in the primary auditory
cortex of naïve (untrained) and trained
30 rats (A). They found that the cortical
20 t erritory containing neurons tuned to
frequencies near 6 kHz (green zones) had
10 expanded in the trained rats, relative to
naïve rats, whereas the territory dedicated
0 to higher frequencies (yellow to red shad-
1 2 4 8 16 32 64 ing) had shrunk. A quantitative summary is
Best frequency (kHz) shown in (B). [From Weinberger, 2007]
Cortical area (%) 30
Cortical area (%)
20
10
0
1 2 4 8 16 32 64
Best frequency (kHz)
in the map. Analogous forms of learning-related plasticity have been observed in the
primary somatosensory and visual cortices.
The mechanisms underlying the overrepresentation of conditioned stimuli in
the sensory cortex remain unclear. Learning-related changes in the auditory map re-
quire acetylcholine release in the neocortex, and repeatedly pairing acetylcholine
release with stimulus presentation is sufficient to generate plasticity. However, it is
not known whether the changes in functional organization involve axonal sprouting
and spine turnover or just changes in the strength of existing synapses. Some learning-
related changes in the sensory cortex occur within an hour of training and are there-
fore unlikely to require the sprouting of long axon branches. However, other changes
develop over the course of several days, suggesting that they may involve axon sprout-
ing. In any case, the data clearly show that neurons in the primary sensory cortices
are not simply providing the raw material for learning and memory; they are part of
the circuitry that is involved in learning and memory.
Motor Cortex Plasticity
The primary motor cortex (see Figure 1.14) exhibits learning-related plasticity very
similar to that observed in the sensory cortex. This plasticity can be demonstrated by
comparing cortical motor maps (showing which neurons are involved in which
movements) between animals that learned a complex motor skill and animals that
performed much simpler movements. In one experiment, adult rats were trained
to reach through a small opening in their cage and grasp a slowly moving food pellet.
At the end of the training period, the motor cortex of the rats was mapped and
compared to the motor cortex of control rats that simply pressed a lever to get food.
Remarkably, the rats that had learned the skilled movement dedicated significantly
more of their motor cortex to movements of the forelimb digits and wrist, which were
essential to grasping the food, than the rats that had simply pressed the lever repeat-
edly (Figure 3.19). The rats in the skilled learning group didn’t just expand the areas
required for the skilled movements; they also shrank the areas devoted to movements
90 Chapter 3 Neuronal Plasticity
A Skilled reaching task B Unskilled reaching task
Figure 3.19 Training can alter the motor Forelimb Movements Hindlimb area
cortex map.Adult rats were trained to area elicited in: 1 mm
reach for moving food pellets (a task C
requiring skill) or simply press a bar for 0.4 Toes Toes
food (an unskilled task). Shortly after the Wrist
last training session, the experimenters Elbows/shoulder
used electrical microstimulation to map Head/neck
the motor cortex in both groups of ani- Hindlimb
mals. They found that the representation Non-responsive
of the forelimb’s wrist and toes was larger
in the skilled group (A) than the unskilled 5 Wrist 4 Elbow/shoulder
group (B). The bar graphs in (C) show
thatthe wrist and toe representations Area (mm2)
expanded at the expense of the elbow Area (mm2)
and shoulder representation. [After Area (mm2)
Kleimet al., 2002]
0 0 0
Skilled Unskilled Skilled Unskilled Skilled Unskilled
of the elbow and shoulder, which are much less important for grasping a food pellet.
The general conclusion is that learning a skilled motor task leads to an expansion of
the cortical territory that is in involved in controlling the learned movements (at the
expense of other areas). By contrast, repetitive performance of an easy, unskilled task
causes little or no plasticity.
Similar findings have been reported for humans. For example, the hand region of
the cortical motor map is expanded in highly skilled pianists and violinists, relative
to nonmusicians. Because violinists make faster, more differentiated movements
with the fingers of their left hand than their right, you would expect violinists to show
a disproportionate enlargement of the hand area in the right motor cortex, which
controls movements of the left hand. Indeed, they do! Given that fine motor control
requires detailed sensory feedback, you might also expect the sensory representation
of the hand to be enlarged in skilled violinists. Again, this prediction has been con-
firmed. These observations raise an interesting question: did the sensory and motor
maps change as a result of extensive practice, or did the musicians become good at
their craft because their cortex was unusual to begin with? A tentative answer to this
question comes from the observation that the degree of cortical sensory and motor
enlargement tends to correlate with the number of years of musical training. Thus,
the data suggest that, the more you practice, the more the cortical areas related to the
practiced skill will grow.
An interesting demonstration of motor cortex plasticity involved monkeys whose
motor cortex had been chronically implanted with electrodes at two different loca-
tions, sites A and B. At the beginning of the experiment, electrical stimulation at
these two sites evoked two very different movements. The experimenters then stimu-
lated site B every time the neurons at site A fired an action potential. After two days
of such stimulation, while the monkey was moving freely around its cage, the experi-
menters again stimulated sites A and B separately. They found that the evoked move-
ments were now quite similar at the two sites, with the movements elicited by
stimulation at site B having changed more than those evoked at site A. This observa-
tion suggests that the two days of stimulation caused the neurons at site A to project
more strongly to the neurons at site B, just as Hebb’s rule would predict (because
the neurons at sites A and B were active simultaneously). More work is needed to
determine whether this plasticity involves the sprouting of new connections or just
Can Experiences Rewire the Brain? 91
THERAPIES
Box 3.3 Brain–Machine Interfaces
Brain–machine interfaces (BMIs) make it possible for paralyzed subject’s own movements to figure out how neural activity
subjects to control machines, such as a wheelchair or a robot in the motor cortex correlates with movement. To solve this
arm, without contracting a muscle. The earliest studies on problem, researchers typically ask the subject to imagine
BMIs were conducted on non-humans. After implanting elec- moving the robot arm. Because neural activity during imag-
trodes in the motor cortex of monkeys or rats, researchers ined movements correlates with activity during actual move-
recorded the activity of multiple neurons as the animals per- ments, researchers can use those correlations to construct the
formed diverse movements. Using a computer to identify cor- decoder. This is great, but there’s another problem, namely,
relations between movement parameters and patterns of that the implanted electrodes tend to move within the brain
neural activity, the researchers built a “decoder” that can infer over the course of several days or weeks, which means that the
the subject’s intended movements from the neural activity and population of recorded neurons changes slowly over time.
do so in real time. The experimenters then restricted the animal’s This “population drift” problem can be solved by regularly
own movements, while continuing to record the neural activity, recalibrating the decoder using the imagined movement
and used the decoder to infer the subject’s intended move- method. Alternatively, or in addition, experimenters can pro-
ments. Another computer then translated the inferred inten- vide the decoder with feedback about how well it decoded
tions into commands that drive the robot arm. the subject’s intended movements. If the movement of the
robot arm did not match the intention, the decoder’s algo-
Using this approach to help paralyzed humans is difficult rithm is altered; when performance is good, no changes are
because their inability to move means that you cannot use the
made. In this manner, the decoder learns to optimize its
Figure b3.3 Paralyzed but able to control a robot arm.A paraplegic performance.
woman learned to control a robotic arm using a brain–machine inter-
face. Two sets of electrodes implanted in the motor cortex are con- What does this have to do with neuronal plasticity?
nected to a computer, which translates the subject’s intended The answer is that the computer is not the only one that
movements into signals that control the robot arm. Both the woman learns. The paralyzed subject also learns to use the BMI.
and the computer use error signals to improve their performance. In fact, providing a subject with visual feedback on the
Ittook several months of regular practice, but now the subject can success or failure of their attempts to use the robot arm
feedherself a chocolate bar. Video at http://youtube/76lIQtE8oDY. greatly improves their rate of progress in controlling that
[Image courtesy of University of Pittsburgh Medical Center] arm. Over several days, such feedback leads to changes
in the “movement tuning” of the recorded neurons. The
mechanisms underlying this form of brain plasticity
remain unclear but probably include the kind of remap-
ping you learned about in this chapter.
Although changes in the brain might force the de-
coder to adjust its algorithm, and changes in the decoder
might force the brain to tweak its own activity, brain
and decoder tend to learn in concert with one another,
coadapting their internal mechanisms to optimize per-
formance. The effectiveness of this approach is best
illustrated by a paraplegic subject who learned to feed
herself with a robotic arm after just a few months of
diligent practice (Figure b3.3). Remarking on her own
progress, she said, “I used to have to think, up, clockwise,
down, forward, back. . . . Now I just look at the target, and
Hector [the arm] goes there.”
changes in synaptic strength. Either way, the finding implies that the internal wiring
of the motor cortex can be modified by electrical stimulation. Researchers are now
exploring how this knowledge might be used to develop better neural prostheses,
which allow patients with paralyzed limbs to control robot arms or other machines
(see Box 3.3).
BRAIN EXERCISE
How would you expect the neocortex of professional baseball players to differ
from that of people who have never played baseball? How might you determine
whether those differences are the result of experience (learning)?
92 Chapter 3 Neuronal Plasticity
A Total brain Cerebral cortex
1680 (mg) (mg)
680
Figure 3.20 Environmental enrichment 1644 644
effects.Total brain and cerebral cortex
mass are increased in rats that grew up in 1600 600
an enriched environment with access to Deprived Enriched Deprived Enriched
various “toys,” interactions with other rats,
and daily maze training (A). Rats in the B
“deprived” comparison group are housed
individually in standard rat cages (no toys). Synapses Neurons Synapses/neurons
Panel (B) demonstrates that environmental (per mm3) (per cm3)
enrichment decreases cortical neuron 100 10,000
density without changing synapse density.
Therefore, the number of synapses per .8 80 8,000
neuron is increased. Rats housed with .6 60 6,000
other rats but without toys or daily training
(the “social” condition) exhibit only weak .4 40 4,000
enrichment effects. [After Rosenzweig et al., Deprived Social Enriched Deprived Social Enriched Deprived Social Enriched
1962, and Turner and Greenough, 1985]
3.6 How Does Experience Affect Brain
andCortexSize?
One form of neuronal plasticity that we have not yet discussed is experience-depen-
dent growth of the entire brain or major brain regions. To demonstrate this kind of
large-scale plasticity, Mark Rosenzweig and his colleagues in the 1960s studied the
brains of laboratory rats that were housed in two very different environments. Some
rats were housed by themselves in small cages without toys or opportunities for exer-
cise. A second group of rats was given daily maze training and housed in roomy cages
with social companions, toys, ladders, and running wheels. Two months later the rats
were sacrificed. As Rosenzweig discovered, the rats living in the impoverished envir-
onment ended up having a significantly smaller neocortex than the rats that had been
living in the enriched conditions. Subsequent studies using electron microscopy re-
vealed that the visual cortex in the enriched rats contains larger neurons, a larger
number of synapses per neuron, and more glial cells (Figure 3.20).
Environmental enrichment or deprivation can also affect the size of human brains.
Most dramatically, the brains of orphans who were institutionalized under atrociously
deprived conditions in Romania (under the Ceaușescu regime) were 16% smaller, on
average, than control brains. The orphans also suffered from a wide variety of cognitive
defects. Whether these neural and behavioral effects of developmental deprivation can
be reversed, at least to some extent, remains unclear. Similarly, it remains unclear why
the effects of deprivation are more severe in some children than others, even when the
deprivation was equally severe. Some of this residual variation is likely genetic.
BRAIN EXERCISE
Do you think laboratory rats and mice should be housed in enriched or deprived
conditions (or somewhere in between)? What might be some pros and cons?
3.7 Does Neural Plasticity Cause Learning
andMemory?
You have now learned that even adult brains exhibit diverse forms of neuronal plasti-
city. They may change the strength of some existing synapses, form novel connec-
tions or eliminate established ones, change the number of neurons representing
Does Neural Plasticity Cause Learning andMemory? 93
a particular function, and grow or shrink in absolute size. This neuronal plasticity
often accompanies the learning of new information or new skills, confirming what
Ramón y Cajal and others had suspected back in the late 1800s. Based on a mountain
of evidence, neuroscientists now know that experience-dependent neuronal plasticity
may be found in many different brain regions, including the hippocampus, sensory
and motor cortices, and the cerebellum. Although plasticity in these brain areas
tends to accompany learning and memory, providing convincing evidence that the
neural changes cause the learning or the memories is difficult.
To establish a causal link between learning and neural plasticity, researchers can
ask whether preventing the plasticity impairs the learning. Such loss-of-function
experiments have shown, for example, that CREB is necessary for long-term sen-
sitization in Aplysia and that blocking LTP in the hippocampus can interfere with
memory. It is important to note, however, that negative results in such experiments
do not disprove a causal link because other mechanisms may compensate. For ex-
ample, the auditory cortex is probably involved in tone conditioning, even though
animals with lesions of the auditory cortex can still associate a tone with noxious
stimuli, most likely using plasticity in some subcortical circuits (see Chapter 14).
Gain-of-function experiments that determine whether boosting plasticity is
sufficient to enhance learning and memory can also provide evidence for causal
links; but negative results are, once again, inconclusive. For example, you learned
that injecting phosphorylated CREB into an Aplysia sensory neuron suffices to induce
long-term sensitization at its synapse onto the gill motor neuron. Combined with the
aforementioned loss-of-function results, these data provide compelling evidence that
CREB is causally linked to long-term sensitization. However, negative results in such
gain-of-function experiments would be difficult to interpret. The manipulation might
have targeted an insufficient number of neurons, or additional mechanisms may have
to be altered before a gain-of-function becomes evident. Even positive results must
be interpreted with care. For example, boosting NMDA receptor function may facili-
tate learning and memory, but it may do so through a variety of mechanisms, some of
which may have nothing to do with LTP. Boosting NMDA receptor function might,
for instance, make an animal more sensitive to noxious stimuli, which would make
the animal more likely to learn associations with noxious stimuli; but we would not
be able to conclude that NMDA receptors are causally linked to memory. The differ-
ence in sensitivity would be a potential confound.
One other important way to test for causal links between neuronal plasticity
and memory is to ask whether the degree of plasticity correlates with the amount of
learning animals exhibit. If the correlation is strong, then a causal link is probable.
For instance, the hypothesis that spine turnover is causally linked to song learning
in birds is bolstered by the observation that changes in the rate of spine turnover cor-
relate with the fidelity of the song imitations. Similarly, the hypothesis that changes
in the tonotopic map of the auditory cortex are causally linked to learning and
memory is supported by data indicating that the amount of map plasticity correlates
with the strength of a rat’s memory for the learned association. In the motor cortex,
too, learning a difficult motor skill elicits more plasticity than learning a simple task
and then performing it repeatedly. Given such correlations between learning and
plasticity, a causal link between the two is highly probable. Finally, it is worth noting
that both learning and neural plasticity are greater when the information being
learned is important to the animal. As you know, it is difficult to learn new informa-
tion if it seems irrelevant. The larger lesson here is this: if you want to learn something
so well that it alters your neural circuitry, then try to figure out why the information
is relevant and important to you.
BRAIN EXERCISE
In your view, what would be the best possible experiment to test whether
extensive violin training changes the cortical representation for movements
ofthe left hand? What would be some problems with this ideal experiment?
94 Chapter 3 Neuronal Plasticity
SUMMARY
Section 3.1 - The mechanisms underlying synaptic plas- Section 3.3 - The best studied form of synaptic weakening
ticity have been studied extensively in the marine snail is cerebellar long-term depression (LTD), which involves
Aplysia californica. metabotropic glutamate receptors and the removal of
AMPA receptors from the postsynaptic membrane.
•Sensitization of the gill withdrawal reflex involves the
release of serotonin onto presynaptic terminals in a Section 3.4 - Synaptic scaling refers to neurons increasing
simple sensorimotor circuit. Activation of the serotonin the average strength of all their synaptic inputs after they
receptors triggers an intracellular signaling cascade have been relatively inactive and decreasing the strength
that ultimately broadens the presynaptic action po- of those inputs when neurons have been very active.
tentials, which then boosts transmitter release and
increases EPSP amplitude. Section 3.5 - Adult brains may rewire themselves as a
result of experience.
•Long-term sensitization requires new protein synthe-
sis. A crucial factor is the phosphorylation of CREB, •Dendritic spines may sprout or disappear in adult
which regulates the transcription of multiple genes brains, and changes in the rate of spine turnover may
involved in making changes in synaptic strength more correlate with learning. Axonal branches may also
persistent. sprout as animals learn.
Section 3.2 - Synapse strengthening (potentiation) in •As animals learn important sensory information, the
mammals is commonly induced by high-frequency repeti- cortical territory representing that information tends
tive (tetanic) stimulation of an input pathway. to expand at the expense of territory representing less
important information.
•Long-term potentiation (LTP) is characterized by an
increase in EPSP amplitude that persists for several •As animals learn a new motor skill, portions of the
hours (or more) after the end of the tetanic stimulus. motor cortex related to that skill tend to expand at the
It is most often studied in hippocampal slices. expense of other areas. The more difficult the skill,
the greater the expansion.
•LTP is associative when it obeys Hebb’s rule, which
states that synapses should only be strengthened Section 3.6 - Animals living in an enriched environment
when presynaptic activity is accompanied by post- tend to have a larger neocortex than deprived animals. The
synaptic depolarization, which may trigger action enlargement is due mainly to the growth of individual
potentials. neurons and the addition of glial cells.
•Glutamate can open NMDA receptors only when the Section 3.7 - Establishing a causal link between neural
postsynaptic membrane is depolarized enough to phenomena and behavior is best done by combining gain-
remove their magnesium block. NMDA receptor acti- of-function and loss-of-function experiments with cor-
vation triggers an intracellular cascade that ultimately relative data.
causes more AMPA receptors to be inserted into the
postsynaptic membrane. It can also stimulate synapse Box 3.1 - It is easy to underestimate invertebrates, both in
growth. their cognitive capacities and in their impact on neuroscience.
•The stabilization of LTP requires protein synthesis. Box 3.2 - The hippocampus probably plays similar roles in
Dendritic protein synthesis and synaptic tagging are spatial and episodic memory in both humans and rodents,
two possible mechanisms for ensuring that the newly but this idea remains debatable.
synthesized proteins strengthen only those synapses
that contributed to the postsynaptic depolarization Box 3.3 - Paralyzed humans can control a robot arm
(as required by Hebb’s rule). through a brain–machine interface. Optimal performance
is reached when both the machine and the human learn
through feedback about motor performance.
KEY TERMS cyclic adenosine monophosphate ubiquitin hydrolase 76
(cAMP)73 heterosynaptic 76
Aplysia californica 72 hom*osynaptic 76
gill withdrawal reflex 72 protein kinase A (PKA) 73 tetanic stimulation 77
semi-intact preparation 72 post-tetanic potentiation 77
sensitization 73 cAMP response element binding
adenylate cyclase 73 protein(CREB) 75
Additional Readings 95
perforant path 77 synaptic tagging hypothesis 83 protein kinase C 85
dentate gyrus 77 dendritic protein synthesis 83 spike timing-dependent plasticity 86
long-term potentiation 79 activity-related cytoskeletal protein synaptic scaling 86
brain slice 79 excitatory postsynaptic current
Hebb’s rule 80 (ARC) 83
magnesium block 80 zeta inhibitory peptide (ZIP) 84 (EPSC)86
calmodulin 81 protein kinase M-zeta (PKM-zeta) 84 spine turnover 87
calcium/calmodulin kinase II epigenetic change 84 tonotopic map 88
long-term depression (LTD) 84 motor map 89
(CaMKII) 81 parallel fiber 85 loss-of-function experiment 93
caged glutamate 82 climbing fiber 85 gain-of-function experiment 93
ADDITIONAL READINGS
3.1 - Synaptic Plasticity in Aplysia 3.3 - Synaptic Weakening
Kandel ER. 2001. The molecular biology of memory storage: a Gao Z, van Beugen BJ, De Zeeuw CI. 2012. Distributed syn-
dialogue between genes and synapses. Science 294:1030–1038. ergistic plasticity and cerebellar learning. Nat Rev Neurosci
13:619–635.
3.2 - Synaptic Plasticity in Mammals
Cooke SF, Bliss TVP. 2006. Plasticity in the human central 3.4 - Synaptic Scaling
nervous system. Brain 129:1659–1673. Abbott LF, Nelson SB. 2000. Synaptic plasticity: taming the
Cooper S. 2005. Donald O. Hebb’s synapse and learning beast. Nat Neurosci 3(Suppl):1178–1183.
rule: a history and commentary. Neurosci Biobehav Rev
28:851–874. 3.5 - Synaptic Rewiring
Day JJ, Sweatt JD. 2011. Cognitive neuroepigenetics: a role Barnes SJ, Finnerty GT. 2010. Sensory experience and cortical
for epigenetic mechanisms in learning and memory. Neuro- rewiring. Neuroscientist 16:186–198.
biol Learn Mem 96:2–12. Bieszczad KM, Weinberger NM. 2010. Representational
Derkach V, Barria A, Soderling TR. 1999. Ca2+/calmodulin- gain in cortical area underlies increase of memory strength.
kinase II enhances channel conductance of alpha-amino- Proc Natl Acad Sci U S A 107:3793–3798.
3-hydroxy-5-methyl-4-isoxazolepropionate type glutamate Galván VV, Weinberger NM. 2002. Long-term consolidation
receptors. Proc Natl Acad Sci U S A 96:3269–3274. and retention of learning-induced tuning plasticity in the audi-
Gustafsson B, Wigström H, Abraham WC, Huang YY. 1987. tory cortex of the guinea pig. Neurobiol Learn Mem 77:78–108.
Long-term potentiation in the hippocampus using depolar- Hihara S, Notoya T, Tanaka M, Ichinose S, Ojima H,
izing current pulses as the conditioning stimulus to single Obayashi S, et al. 2006. Extension of corticocortical afferents
volley synaptic potentials. J Neurosci 7:774–780. into the anterior bank of the intraparietal sulcus by tool-use
Korb E, Finkbeiner S. 2011. Arc in synaptic plasticity: from training in adult monkeys. Neuropsychologia 44:2636–2646.
gene to behavior. Trends Neurosci 34:591–598. Münte TF, Altenmüller E, Jäncke L. 2002. The musician’s brain
Lisman J, Yasuda R, Raghavachari S. 2012. Mechanisms of as a model of neuroplasticity. Nat Rev Neurosci 3:473–478.
CaMKII action in long-term potentiation. Nat Rev Neurosci Rosenkranz K, Williamon A, Rothwell JC. 2007. Motorcortical
13:169–182. excitability and synaptic plasticity is enhanced in professional
Martin KC, Zukin RS. 2006. RNA trafficking and local protein musicians. J Neurosci 27:5200–5206.
synthesis in dendrites: an overview. J Neurosci 26:7131–7134.
Martin KC, Kosik KS. 2002. Synaptic tagging—who’s it? Nat 3.6 - Changes in Brain and Cortex Size
Rev Neurosci 3:813–820. Anderson BJ. 2011. Plasticity of gray matter volume: the cellu-
Murakoshi H, Yasuda R. 2012. Postsynaptic signaling during lar and synaptic plasticity that underlies volumetric change.
plasticity of dendritic spines. Trends Neurosci 35:135–143. Dev Psychobiol 53:456–465.
Sacktor TC. 2011. How does PKMζ maintain long-term Sheridan M, Drury S, Mclaughlin K, Almas A. 2010. Early
memory? Nat Rev Neurosci 12:9–15. institutionalization: neurobiological consequences and ge-
Volk LJ, Bachman JL, Johnson R, Yu Y, Huganir RL. 2013. netic modifiers. Neuropsychol Rev 20:414–429.
PKM-ζ is not required for hippocampal synaptic plasticity,
learning and memory. Nature 493:420–423. 3.7 - From Correlation to Causality
Destexhe A, Marder E. 2004. Plasticity in single neuron and
circuit computations. Nature 431:789–795.
96 Chapter 3 Neuronal Plasticity
Tang Y, Shimizu E, Tsien JZ. 2001. Do “smart” mice feel Nadel L, Hoscheidt S, Ryan LR. 2013. Spatial cognition and
more pain, or are they just better learners? Nat Neurosci the hippocampus: the anterior-posterior axis. J Cogn Neuro-
4:453–453. sci 25:22–28.
Orsborn AL, Carmena JM. 2013. Creating new functional
Boxes circuits for action via brain-machine interfaces. Front Comput
Burne T, Scott E, van Swinderen B, Hilliard M, Reinhard J, Neurosci 7:157.
Claudianos C, et al. 2011. Big ideas for small brains: what Velliste M, Perel S, Spalding MC, Whitford AS, Schwartz AB.
can psychiatry learn from worms, flies, bees and fish? Mol 2008. Cortical control of a prosthetic arm for self-feeding.
Psychiatry 16:7–16. Nature 453:1098–1101.
Developing a CHAPTER
Nervous System
4
CONTENTS FEATURES
4.1 Where in the Embryo Does the Nervous System RESEARCH METHODS
Originate?
Box 4.1 In Situ Hybridization
4.2 How Does the Neural Tube Get Subdivided?
EVOLUTION IN ACTION
4.3 Where Do Neurons Come From? A
4.4 How Do Axons Find Their Targets? Box 4.2 Hox Genes in Evolution
NEUROLOGICAL DBISORDERS
Box 4.3 Drugs and a Baby’s Brain
4.5 How Do Synapses Form?
4.6 How Can a Neural Circuit Be Fine-Tuned?
4.7 What Are the Major Themes of Neural Development?
Fruit fly C Mammal D
1 2 3* 3* 3* 4 5 6* 6 1 2 3 4 5 6 7 8 9 10 1112 13 Ev/Hx
A
7 8 9/13
eve(Ev/Hx) B
C
D
Urochordate 11
1 10 13
2 4
Nematode 9 12
Ev/Hx
A B 51 6/8 6/8 Actin
Filopodia filaments
Sensory 9 9 C Ev/Hx Cephalochordate Ev/Hx
neurons
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Platyhelminth 3/ftx? Sea urchin
8
1 5 6 7 8 9/10 10/13 32 1 Ev/Hx
4
97
1 2 3 4 5 6/8 9/14 Ev/Hx
98 Chapter 4 Developing a Nervous System
32 cell stage Although it might appear that every neuron in a complex
nervous system is connected to every other neuron, a
closer look reveals that real neurons tend to have rather
specific connections with select groups of other cells. Just
as neurons differ in size, shape, transmitters, and ion chan-
nel types, so they differ in axonal connections. Further-
more, neurons with similar connections tend to cluster
together, forming discrete brain areas, laminae, or nuclei.
Thus, real nervous systems are heterogeneous (spatially
All cells β-catenin variable) rather than hom*ogeneous. But how does struc-
tural and functional heterogeneity arise during develop-
ment? Given that development starts with a single fertilized
egg cell, how do some cells of the growing embryo get spec-
ified to become the nervous system? How do the various
cells in the nervous system become different from one
another? How does each neuron “know” to which other
neurons it should be connected?
On one level, the answer is relatively simple. As a fertil-
ized egg divides again and again, its daughter cells come to
express different sets of genes, which causes the cells to
vary in size, shape, and other properties (Figure 4.1). This
only begs the question, however. What causes different
daughter cells to express different genes, given that they all
contain the same DNA? The answer is that different cells
contain different transcription factors, which contain
specific DNA-binding domains that, when bound, promote
(or repress) the expression of genes close to the binding
site. In other words, they recognize specific DNA se-
quences, bind to them, and then regulate gene transcrip-
tion at those selected locations (you may recall that CREB,
which we discussed in Chapter 3, is one such transcription
factor). Because transcription factors regulate many genes,
some of which in turn encode other transcription factors
or proteins that mediate cell–cell interactions, differences
in transcription factor expression usually translate into
Blastocyst long-lasting differences in how the cells behave, both intra-
cellularly and in their interactions with other cells. Thus,
differences in transcription factor expression tend to pro-
Figure 4.1 Breaking symmetry in embryos.In very young sea anem- duce differences in cell fate.
one embryos, every cell expresses beta-catenin (β-catenin; top). As Of course, this still doesn’t answer the more fundamen-
development proceeds, β-catenin expression becomes restricted to tal question: what causes different cells to express different
one side of the embryo (the future endoderm). This change in gene transcription factors? For some invertebrates, the answer
expression is an example of symmetry breaking because the embryo is lies with mom, who makes her eggs so that they already
less symmetrical after the change. [From Wikramanayake et al., 2003]
contain an asymmetric arrangement of transcription fac-
tors. In mammals, both the mother and the father’s sperm cooperate to distribute
transcription factors heterogeneously within the fertilized egg. This need not con-
cern us here. What is important is that spatial differences in transcription factor ex-
pression are crucial to the early formation of the nervous system.
4.1 Where in the Embryo Does the Nervous
SystemOriginate?
Five to six days after fertilization, a human egg has grown into a hollow clump of cells
that would fit comfortably on the head of a pin. At or before this blastocyst stage,
embryonic cells that accidentally get separated from the others can form a complete
embryo, an identical twin. After the blastocyst stage, twinning is rare because the
Where in the Embryo Does the Nervous SystemOriginate? 99
embryonic cells become more specialized. In particular, the cells of the blastocyst
rearrange themselves during gastrulation to form three distinct germ layers: namely,
ectoderm, endoderm, and, sandwiched between them, mesoderm (Figure 4.2). The
ectoderm is of special interest to neurobiologists because its cells differentiate into
two seemingly very different tissues: epidermis (skin) and nervous system. It may
seem strange that skin and nervous system are developmentally so closely related,
but the earliest animals probably had their entire nervous system located within the
skin. Centralized brains and spinal cords evolved later (see Chapter 16).
Induction of the Nervous System
Why do some ectodermal cells develop into skin while others form the nervous
system? A key experiment addressing this question was performed by Hilde Mangold
and her dissertation advisor Hans Spemann in the 1920s (Figure 4.3). They took a
piece of mesoderm called the dorsal blastopore lip from the embryo of a white
(lightly pigmented) amphibian just after the blastocyst stage and transplanted it into
a darkly pigmented amphibian embryo of the same age. This modified embryo had
two blastopore lips, one on each side of the embryo, and it developed into Siamese
twins, joined at the belly but with two complete nervous systems (Figure 4.3). Cru-
cially, Mangold noticed that the cells in both nervous systems were darkly pigmented,
Zygote Gastrula
Ectoderm Host’s pigmented
Epidermis and dorsal blastopore lip
nervous system
Blastocyst
Transplanted albino
dorsal blastopore lip
Siamese Twins
Gastrula
Mesoderm Endoderm
Muscle, bone
and cartilage Lining of gut
and lungs
Yolk sac Both pigmented
Figure 4.2 From zygote to gastrula.The zygote is a fertilized egg. Figure 4.3 The Mangold–Spemann “organizer” experiment.Mangold
The blastocyst (shown in cross section) is a hollow ball of cells. The and Spemann cut the dorsal blastopore lip out of an albino amphibian
gastrula (also sectioned) contains the three germ layers—endoderm, embryo and transplanted it into the ventral pole of a pigmented
mesoderm, and ectoderm—suspended as a flat sheet between g astrula. When such a gastrula grows up, it produces Siamese twins
twofluid-filled spaces (blue and yellow). The ectoderm develops (bottom). Because both twins are pigmented, the transplanted blasto-
mainly into the epidermis (skin) and nervous system. The endoderm pore lip must have “induced” its host to form the second embryo. This
is fated to become the lining of the gastrointestinal tract and lungs. experiment (originally performed on salamanders) led to the idea that
The mesoderm is sandwiched between the other two germ layers the dorsal blastopore lip is an “organizer” that controls the fate of
and spreads laterally to surround the yolk sac. It gives rise to muscle, surrounding tissue. [From De Robertis and Kuroda, 2004]
c onnective tissue, and red bloodcells.
100 Chapter 4 Developing a Nervous System
RESEARCH METHODS
Box 4.1 In Situ Hybridization
The method of in situ hybridization (ISH) reveals when and template provided by the other DNA strand (Figure b4.1). Cru-
where specific genes are expressed. It is a widely used tech- cially, some of the free nucleotides in the solution are conju-
nique that has played an exceptionally important role in devel- gated to a molecule that researchers can later visualize. Often
opmental neurobiology. Although the method is complex, its the labeled nucleotide is deoxyuridine triphosphate (dUTP)
basic principles are straightforward. The process begins with bound to biotin or digoxygenin. When the DNA incorporating
the construction of a “probe” for the gene of interest. One way these labeled nucleotides is separated into single strands
to make such a probe is to intentionally damage the isolated (denatured), the labeled single strands become the desired
gene’s DNA in a few places with the enzyme Deoxyribonucle- “probe” that can bind to other samples of denatured DNA or
ase I (DNase-I). This “nicked” DNA is then incubated with DNA to RNA.
Polymerase I (Pol-I), which replaces any removed nucleotides To determine when and where in an embryo genes are
with new ones from the surrounding solution, based on the transcribed, ISH probes must be brought into contact with
the tissue’s RNA. This is often ac-
complished by fixing the embryo
(often by immersing it in parafor-
Gene of interest RNA in maldehyde) and slicing it into thin
fixed tissue sections that are then mounted
onto glass slides. Alternatively,
a whole embryo can be treated
DNase-1 with detergents and/or proteases
Nick Labeled Hybridization to poke them full of holes through
translation nucleotides which ISH probes can pass. The
latter method is called whole mount
in situ hybridization because the
Pol-1 ISH probe tissue is placed (mounted) in the
Denaturation ISH reaction apparatus without
having been sliced. Sectioned
or not, the tissue containing the
probes is processed through sev-
eral solutions to wash away any
Probe probe that is not tightly bound
visualization to the intended target. Then the
probe is visualized with antibod-
Fluorescent ies that are bound to fluorescent
antibodies molecules. These antibodies can
be seen through microscopes
with special light sources and
optical filters; they end up glow-
ing in the dark (typically in shades
of red, yellow, or blue). Alterna-
tively, ISH probes can be subjected
Figure b4.1 The in situ hybridization technique.To make an RNA probe, one takes DNA from to a series of chemical reactions
the gene of interest and substitutes unlabeled nucleotides with nucleotides labeled with biotin that generate a colored product
or digoxygenin (yellow circles). The double stranded DNA is then denatured to generate a single- visible through a standard light
stranded probe, which is brought into contact with RNA in the tissue of interest (lightly shaded microscope. Either way, you end
ovals). Under the proper conditions, the probe binds selectively to (hybridizes with) any RNA up with a colorful image, where
thathas the complementary nucleotide sequence. After the hybridization, the labeled probe the presence of the color indi-
isexposed to antibodies that fluoresce at a specific wavelength (red circles linked to blue cates the presence of the tran-
Y-shaped structures). scribed RNA.
implying that they must have developed from cells that came from the “host” embryo.
The transplanted, lightly pigmented cells developed into structures that were adjacent
to the second nervous system. Based on these observations, Mangold and Spemann
hypothesized that cells of the dorsal blastopore lip emit some sort of signal that
Where in the Embryo Does the Nervous SystemOriginate? 101
induces neighboring cells to develop into a nervous system. This conclusion raised
many questions about the cellular and molecular mechanisms underlying nervous
system induction.
The search for the neural inducer (organizer) molecule hypothesized by Mangold
and Spemann advanced significantly in the 1990s when experimenters used in situ
hybridization (Box 4.1: In Situ Hybridization) and other molecular techniques to
identify several genes that are expressed selectively in the dorsal blastopore lip. One
of these genes is chordin. When injected into embryos, molecules of the chordin pro-
tein cause ectodermal cells to differentiate into neural tissue. Chordin does this by
inhibiting bone morphogenetic protein (BMP), which is secreted by cells in the
ventral portion of the embryo (Figure 4.4). When BMP molecules diffuse away from
their ventral source and bind to the ectoderm, they cause ectodermal cells to become
skin. However, as BMP molecules approach the dorsal blastopore lip, they are inacti-
vated by chordin and other BMP antagonists (other blockers of BMP function). What
do ectodermal cells become if they do not receive the BMP signal? They become the
nervous system.
The discovery that inhibiting BMP causes the ectoderm
to develop into neural tissue was surprising because most
people had expected that becoming the nervous system would Amnion
be something special, not the ectoderm’s “default” mode. (cut edge) Neural plate - 16 days
Future skin Future nervous
system
Neural fold - 20 days
Dorsal
blastopore
lip (chordin
source)
Somite Notochord
(mesoderm)
BMP Neural tube - 22 days
source
Yolk Neural crest Neural tube
Chordin molecules
BMP molecules
Inactivated BMP
Somite Notochord
(mesoderm)
Figure 4.4 Neural induction.According to a widely accepted model
of neural induction, the dorsal blastopore lip (red) secretes diffusible Figure 4.5 Formation of the neural tube.The neural plate (top) bends
molecules of chordin (orange). When the chordin molecules encounter to form a neural groove (middle), which then closes up to form the
bone morphogenetic protein (BMP) molecules, which are secreted neural tube (bottom). The images on the left depict human embryos
from a source at the other end of the embryo, they prevent the diffus- at16, 20, and 22 days of age. The images on the right illustrate sections
ible BMPs (blue) from binding to the ectoderm. Ectodermal cells that through these embryos (at sites indicated by the light blue section
interact with BMP develop into skin. In contrast, ectodermal cells that planes). These images illustrate mainly the ectoderm, but notochord
are prevented by chordin from interacting with BMP develop into and somites are shown as well. The amnion and other extraembryonic
the nervous system (yellow). Although this model is simplified, it cap- membranes have been cut away. [After images from Marvin Sodicoff]
tures crucial aspects of neural induction, including the observation
that forming the nervous system seems to be the ectoderm’s default
state. The model shown here is for a typical amphibian embryo.
102 Chapter 4 Developing a Nervous System
However, the evidence is pretty clear. Most convincingly, ectodermal cells grown
individually in tissue culture, so that they receive no signals from other cells, adopt
a neural fate. This finding shows that at the very root of nervous system develop-
ment lies not some positive inductive signal, as Mangold and Spemann had thought,
but an inhibitory signal that prevents the alternative outcome of becoming skin.
Forming the Neural Tube
After BMP and its inhibitors have divided the ectoderm into neural and skin-forming
portions, the emerging nervous system is a flat sheet of cells referred to as the neural
plate (Figure 4.5). Soon thereafter the left and right edges of this neural plate lift up,
transforming the plate into a neural groove. As the future skin cells to the left and
right of this groove proliferate, they push the groove’s edges toward the midline until
they meet. At this point, special cell adhesion molecules on the surface of the future
skin cells cause the skin cells on both sides of the neural groove to stick to one an-
other but not to other cells. Neural groove cells express different adhesion molecules,
which make them stick to one another but not to the skin cells. The overall effect of
this selective adhesion is that the neural groove becomes a neural tube that is sepa-
rate from, and covered by, the skin. The neural tube soon closes at the front and at the
rear. It then goes on to form the entire central nervous system, including both brain
and spinal cord.
In addition, so-called neural crest cells migrate away from their original location
right between the skin and the neural plate. They form much of the peripheral nervous
system, including the neurons of the cranial and spinal nerves, the glia associated with
those nerves, the ganglia of the sympathetic nervous system, and the enteric nervous
system. The neural crest also gives rise to a number of non-neural structures, including
skin pigment cells (melanocytes) and much of the skull. This, in brief, is the ectoderm’s
fate. As you can see, a large part of it is destined to become the nervous system.
BRAIN EXERCISE
Why is it fair to say that the ectoderm’s “default” fate is to differentiate into
neural tissue?
4.2 How Does the Neural Tube Get Subdivided?
The study of tissue patterning is the study of how tissue becomes heterogeneous.
For the nervous system, it is the study of how the initially hom*ogeneous neural tube
becomes divided into a complex heterogeneous structure. Although patterning the
neural tube is a complex three-dimensional problem, it can be simplified, at least
initially, by considering rostrocaudal patterning separately from dorsoventral
patterning.
Rostrocaudal Patterning
The spinal cord develops from the caudal portion of the neural tube, whereas the
brain develops from its rostral end. The spinal cord is further subdivided into 31 seg-
ments; and the brain is subdivided into hindbrain, midbrain, and forebrain. Develop-
mental neurobiologists have long wondered how these rostrocaudal divisions of
the central nervous system come into existence. A full answer remains elusive, but
most scientists agree that rostrocaudal neural tube patterning involves molecular
signals that increase in concentration as you go from rostral to caudal along the
neural tube. These molecules caudalize neural tube cells in a concentration-dependent
manner. That is, they cause the affected cells to become caudal, rather than rostral,
neural tissue. One likely candidate for such a caudalizing signal is retinoic acid.
Interfering with retinoic acid signaling prevents caudal brain regions from forming
normally. Conversely, artificial increases in retinoic acid concentration impair the
differentiation of rostral brain regions (Figure 4.6). An important implication of
How Does the Neural Tube Get Subdivided? 103
A Normal RA concentration Hox a-3 Hox b-3
Islet-2
Midbrain Hindbrain Spinal cord r4
Forebrain r5 VI r6 VI
r7
B Increased RA concentration
XII XII
C A model of RA effects
RA concentration Increased RA Figure 4.7 Hox gene expression domains.Shown here are dorsal
Hox concentrationNormal RAviews of the hindbrain from two chick embryos, stained with whole-
mount in situ hybridization to reveal the expression patterns of Hox a-3
(left, purple), Hox b-3 (right, purple stain), and Islet-2 (right, red stain
and arrows). The purple arrows indicate the rostral expression boundar-
ies of the two Hox genes. The labels r4–r7 refer to numbered hindbrain
segments (rhombomeres); and the Roman numerals VI and XII indicate
motor neurons associated with the 6th and 12th cranial nerves, respec-
tively. [From Guidato et al., 2003]
Rostral Spatial position Caudal A Individual hox genes
Rostral
Figure 4.6 Retinoic acid affects brain patterning.Shown in (A) is a Caudal Hox-b2 Hox-a1 Hox-a3 Hox-a4
schematic dorsal view of an embryo that developed with the normal Hox-a2
amount of retinoic acid (RA). Panel (B) shows an embryo that devel-
oped no forebrain because RA levels were artificially increased. B A nested expression pattern
Shown in (C) is a model of RA function, according to which RA con-
centration increases as you proceed caudally. Where RA levels are
high (yellow), the tissue becomes spinal cord. Where they are very
low (purple), the tissue becomes forebrain. At intermediate levels
(blue, green) the neural tube adopts hindbrain and midbrain fates,
respectively. This model predicts that an artificial increase in RA con-
centration (dashed curve) should yield the sort of embryo depicted
in B. [After Maden, 2002]
these findings is that the “default mode” of ectodermal cells Hox-a2
is not just to become neural tissue, but to become rostral Hox-b2
neural tissue (brain).
Hox-a1
Hox-a3
Hox-a4
The Hox Gene Family Rostral Spatial position Caudal
Rostrocaudal neural tube patterning also involves Hox genes.
Hox genes are a family of transcription factors that is highly
conserved across species (Box 4.2: Hox Genes in Evolution).
Individual members of this family are expressed in various Figure 4.8 Nested expression of Hox genes in the hindbrain.
combinations at different rostrocaudal levels of the nervous Shown in (A) are dorsal views of vertebrate hindbrains in which indi-
system (Figure 4.7). Caudal hindbrain segments express many vidual segments are separated by dashed lines. The colors indicate
different Hox genes simultaneously, but more rostral seg- that each Hox gene has a different rostral expression boundary.
Panel(B) summarizes this finding in a single graph to show how
ments express progressively fewer Hox genes (Figure 4.8). thenumber of different Hox genes expressed within a hindbrain
This nested expression pattern suggests that different Hox segment increases as you get to more caudal segments.
genes are activated at different concentrations of a caudaliz-
ing signal, such as retinoic acid. Specifically, Hox genes with the most extensive ex-
pression domain are thought to be activated by very low levels of the caudalizing
104 Chapter 4 Developing a Nervous System
EVOLUTION IN ACTION
Box 4.2 Hox Genes in Evolution
Hox genes were first identified in fruit flies. Ed Lewis and others in life. It determines, for example, whether a body part will grow
discovered that mutations in fruit fly Hox genes cause duplica- antennae or legs.
tions of some body segments and the deletion of others. For By now, hom*ologs of the fruit fly Hox genes have been dis-
example, some Hox gene mutations cause flies to develop legs covered in many animal species (Figure b4.2). As described in
where their antennae should be. Subsequent research revealed the main text, the mammalian Hox genes exhibit a rostrocau-
that fruit flies have multiple Hox genes that all share a conserved dally nested pattern of expression that is similar to that seen in
DNA-binding sequence (called the homeobox). Moreover, the fruit flies. Even more remarkable is that, in both fruit flies and
expression domains of different Hox genes have different ros- rodents, the sequence in which the Hox genes are arranged on
tral boundaries during fruit fly development, which means that the chromosome mirrors (at least roughly) the rostrocaudal
caudal body parts coexpress a larger number of Hox genes sequence of Hox gene expression. The origins and causal im-
than rostral body parts. Additional data reveal that the specific plications of this sequence colinearity remain unclear, but it is
combination of Hox genes that a body part expresses during yet another striking similarity between Hox genes in fruit flies
development determines the identity of that body region later and mammals. Collectively, these similarities suggest that the
Hox gene family has ancient roots,
Fruit fly Mammal likely going back at least to the
last common ancestor of all ani-
1 2 3* 3* 3* 4 5 6* 6 1 2 3 4 5 6 7 8 9 10 1112 13 Ev/Hx mals with bilateral symmetry (the
A Bilateria).
7 8 9/13
eve(Ev/Hx) B The discovery of conserved
C Hox gene expression patterns in
both the body and the central
D nervous system has revitalized an
old debate about whether the
Urochordate 11 central nervous systems of insects
and mammals are hom*ologous.
1 10 13 Are they descended from a cen-
2 4
Nematode 9 12
Ev/Hx
51 6/8 6/8
99 Ev/Hx Cephalochordate
1 2 3 4 5 6 7 8 9 10 11 12 13 14 Ev/Hx
Platyhelminth 3/ftx? Sea urchin tral nervous system that evolved
8 very early in animal phylogeny
1 5 6 7 8 9/10 10/13 32 1 Ev/Hx and was retained continuously
since then, or did the insect and
4 vertebrate central nervous sys-
tems evolve independently of
one another? The conserved Hox
1 2 3 4 5 6/8 9/14 Ev/Hx gene expression pattern is con-
sistent with the former hypoth-
esis. However, it is also possible
Figure b4.2 Hox gene evolution. Individual Hox genes are depicted as thick, colored arrows. The that an ancient set of Hox genes
direction of each arrow indicates the direction in which the gene is transcribed. Most Hox genes became involved in nervous
are numbered in sequence (Hox-1, Hox-2, etc.). hom*ologous genes are given matching names and system patterning long after the
colors across species. The ancestral cluster of Hox genes is thought to have duplicated twice in the genes themselves evolved, and
lineage leading to mammals, producing four Hox clusters (labeled A to D in the Mammal branch). that it did so independently in
Some genes in each cluster were lost. Curiously the presumed ancestral Hox cluster was disbanded multiple lineages. The debate is
in several taxonomic lineages, including flatworms (platyhelminths), round worms (nematodes), ongoing and unlikely to be re-
and tunicates (urochordates). The longest and most orderly Hox cluster is found in the cephalo- solved soon.
chordate Amphioxus. The fruit fly’s Hox cluster is broken into two parts, and some of its genes have
reversed their orientation relative to the ancestral condition. [From Lemons and McGinnis, 2006]
signal, whereas Hox genes expressed only in the caudal hindbrain require high
levels of caudalizing signal. This hypothesis is supported by the finding that artificial
increases in retinoic acid levels cause rostral hindbrain segments to express Hox gene
combinations that are normally found only in more caudal segments. These “caudal-
ized” segments also express non-Hox genes that are typically expressed only in
caudal hindbrain segments of older animals, suggesting that the altered Hox gene
expression pattern permanently alters cell fates.
How Does the Neural Tube Get Subdivided? 105
The details of how Hox genes control cell fate remain a subject of debate. Accord-
ing to the combinatorial Hox code model, a cell’s developmental fate is determined
by the combination of Hox genes expressed within the cell. Alternatively, the poste-
rior prevalence model states that some Hox genes are more important than others.
Specifically, the Hox genes expressed in the more posterior hindbrain segments are
thought to dominate the Hox genes with more anterior expression domains. Accord-
ing to this second model, a cell’s fate depends on the most dominant Hox gene ex-
pressed within the cell. Experiments in which specific Hox genes were “knocked out”
in transgenic mice tend to support the posterior prevalence model, but the matter is
not settled yet. In any case, the data show that Hox genes are essential for rostrocau-
dal patterning of the vertebrate hindbrain and, to some extent, the spinal cord. As
you will see shortly, rostrocaudal patterning in the midbrain and forebrain involves a
different set of transcription factors.
BRAIN EXERCISE
How is the spatial expression of Hox genes linked to levels of retinoic acid in the
embryo (as illustrated in Figure 4.6)?
Dorsoventral Patterning A Dorsal
The central nervous system is also patterned along the dorso- Floor plate Shh protein
ventral axis. In the spinal cord, for example, neurons that send
their axons to muscles lie ventrally; whereas neurons that re- B Ventral
ceive input from sensory nerves are located in the dorsal horn
of the spinal cord. In between these motor and sensory neurons Induced Repressed Late
lie interneurons that connect to other neurons in a manner by Shh by Shh
that varies with their dorsoventral position. As Thomas Jessel
and his collaborators discovered, dorsoventral patterning in Early
the spinal cord involves sonic hedgehog (SHH). This protein
(named after a once-popular video game character) is secreted C
by cells at the neural tube’s ventral midline, which is called the
floor plate. Because SHH diffuses freely away from its ventral Induced Repressed
source, it forms a ventral-to-dorsal concentration gradient by Shh by Shh
within the developing spinal cord (Figure 4.9). This finding sug-
gests that SHH is a ventralizing signal for the embryonic spinal
cord. Indeed, injection of antibodies that block SHH also blocks
the formation of motor neurons that lie ventrally in normal
spinal cords. Conversely, the addition of SHH to embryonic
spinal cords growing in tissue culture induces the cells to
become motor neurons.
Boundary Formation
The ventralizing action of SHH involves two classes of tran- Early Late
scription factors. Members of the first class are induced (turned
on) by SHH. Importantly, different transcription factors within Figure 4.9 Dorsoventral patterning in the spinal cord.The
this class are induced at different concentrations of SHH, which diagram in (A) depicts an embryonic spinal cord in cross section.
causes them to have different dorsal expression boundaries. Floor plate cells secrete sonic hedgehog (SHH), which diffuses
This is analogous to how different Hox genes are induced at dif- away (red circles), setting up an SHH concentration gradient.
ferent concentrations of retinoic acid. In contrast, the second Panel(B) shows the expression of one gene that is induced by
class of transcription factors are repressed (turned off) by SHH high levels of SHH (red), and one that is repressed by SHH (blue;
rather than induced. Moreover, different transcription factors for clarity, each protein is shown on just one side of the spinal
in this group are repressed at different concentrations of SHH, cord although they are expressed on both). Initially these genes
which causes them to have different ventral expression bound- are expressed with spatial overlap (left), but their expression
aries. In general, the expression domains of the SHH-induced d omains gradually become nonoverlapping (right side) because
genes and the SHH-repressed genes are complementary so that these two genes repress each other. Panel (C) is analogous to
as one SHH-induced gene fades out, a SHH-repressed gene’s ex- panel B but illustrates the expression of two other genes that
pression fades in (Figure 4.9). This means that the interactions develop a sharp boundary at a more dorsal location. [After
B riscoe et al. 2000]
106 Chapter 4 Developing a Nervous System
A 100 µm between SHH and its downstream transcription factors
cause the spinal cord to become divided into genetically
Notochord distinct dorsoventral domains.
B
Because the SHH-induced genes generally inhibit
Forebrain Hindbrain the expression of their SHH-repressed complements,
Midbrain Spinal cord and vice versa, individual cells tend to express one
transcription factor or the other but not both. This
Figure 4.10 Development of the midbrain flexure.Shown here is the mutual repression helps young cells make clear deci-
central nervous system at two stages of embryonic development. The sions about their fate. Think of it this way: if cells were
tissue has been cut down the middle and you are looking at it from a to express transcription factors that prompt them to
medial perspective. You can see how the brain bends (flexes) around the become motor neurons simultaneously with transcrip-
anterior end of the notochord, with the midbrain at the apex of the flexure. tion factors that push them toward an interneuron fate,
then the young cells would be “confused” about what to
become. The mutual repression between the different
sets of transcription factors clears up the confusion
and forces the young neurons to choose just one of the
alternative fates. It also leads to a gradual sharpening
of the boundaries between the spinal cord’s dorsoven-
tral compartments.
Midbrain and Forebrain Patterning
Because the neural tube’s rostral end bends dramatically
during development (Figure 4.10), it is not immediately
obvious what is rostral, caudal, dorsal, or ventral in the
brain. However, if you mentally straighten the neural
tube, then you can see that the midbrain lies rostral to
the hindbrain and that the developing telencephalon
lies at the rostral tip of the developing brain (together
with the preoptic area and hypothalamus). You can also
see that SHH is expressed along the ventral edge of
the entire brain, just as it is expressed in the floor of
the hindbrain and spinal cord (Figure 4.11). These data
Pretectum Midbrain
Dorsal
thalamus
Ventral zli Midbrain-
thalamus hindbrain
Pons boundary
Telencephalon
Pituitary Cerebellum
Anterior Hypothalamus
neural ridge
Figure 4.11 Brain patterning.In this Optic Medulla
schematic diagram of an embryonic brain, nerve
sonic hedgehog (SHH) expression is shown Preoptic area
in blue. The spur of SHH expression extend-
ing between the dorsal and ventral thala- Fgf8
mus is called the zona limitans intrathalamica Shh
(zli). Shown in red are two additional
signaling centers, namely, the anterior
neural ridge and the midbrain–hindbrain
boundary. Both secrete (arrows) fibroblast
growth factor 8 (Fgf8), which helps to
pattern the developing brain.
How Does the Neural Tube Get Subdivided? 107
suggest that SHH acts as a ventralizing signal throughout the neural tube, including
both the brain and spinal cord. Indeed, blocking SHH prevents the formation of ven-
tral tissues in all regions of the central nervous system.
The process of mentally straightening the neural tube also reveals that BMP proteins
(yes, the same molecules that induce skin at earlier stages of development) are
expressed dorsally throughout most of the neural tube, including the forebrain. Given
this dorsal expression pattern, it is reasonable to hypothesize that BMP is a dorsalizing
signal that counteracts the ventralizing activity of SHH. Experiments that boost or
block BMP signaling are consistent with this hypothesis. Overall, these data indicate
that dorsoventral patterning is fundamentally similar in the brain and spinal cord.
Rostrocaudal Patterning of the Brain
Rostrocaudal patterning is more complex in the midbrain and forebrain than in the
hindbrain and spinal cord. This greater complexity arises because the brain contains
multiple rostralizing and caudalizing factors secreted by multiple signaling centers.
One of these signaling centers is the midbrain–hindbrain boundary (Figure 4.11).
This ring-shaped region secretes several diffusible transcription factors, including
fibroblast growth factor 8 (FGF8). These signals are crucial for proper midbrain and
cerebellum development.
A second important signaling center develops between the dorsal and ventral
thalamus. This intrathalamic signaling center secretes SHH (Figure 4.11), just as the A
ventral forebrain does. However, SHH in this context does not act as a ventralizing
signal but instead helps to pattern the thalamus. This is a recurring theme in develop-
mental biology: molecules often play different roles in different locations and at
different stages of development.
The third signaling center crucial for brain patterning is the anterior neural Fgf8 Somatosensory
ridge (Figure 4.11). It secretes FGF8, just as the midbrain–hindbrain boundary Fgf8 cortex
does. However, FGF8 from the anterior neural ridge diffuses into the rostral fore-
brain, where it is responsible for rostrocaudal patterning. This was demonstrated in Olfactory
experiments that boosted FGF8 expression in the anterior forebrain, which caused cortex
the somatosensory cortex to develop in laenvealAbsnoofrFmGalFly8 caudal location (Figure 4.12).
That is what you would expect if high promote the development of B
rostral cortical areas. In contrast, reducing FGF8 expression in the rostral forebrain
causes somatosensory cortex to develop in an abnormally rostral location. Again,
this is what you would expect if FGF8 rnoesotcraolritzeexsFgwtfhe8ereceenregbinraeleScroeomdrattteooxse.exnInsporareystshFirGd Fe8x-,
periment, cells at the caudal edge of the
which they do not normally do. This manipulation caused the somcaotortseexnsory cortex
AB C
Fgf8 Somatosensory Fgf8 Fgf8
cortex
Olfactory
cortex
BC
Figure 4.12 Rostrocaudal patterning of neocortex by Fgf8.Shown here are flattened mouse cortices (rostral to the left). Red shading indicates
Fgf8 concentration, the primary somatosensory cortex is shown in purple, and the dashed line provides a rostrocaudal reference. Shown in (A) is
thenormal condition. Panel (B) shows that overexpression of Fgf8 in the rostral neocortex shifts the somatosensory cortex caudally. Conversely,
reducingFFggff88 expression (C) shifts the somatosensory cortFegxfr8ostrally. [After Assimacopoulos et al., 2012]
Olfactory
cortex
108 Chapter 4 Developing a Nervous System
Fgf8 induced to appear in duplicate, once in its normal rostral location
caudally and then again more caudally (Figure 4.13).
S11 S12 In summary, brain and spinal cord patterning depends
on transcription factors that are secreted from various sig-
Fgf8 naling centers and then diffuse away so that each tran-
scription factor forms a distinct concentration gradient.
hl whiskers whiskers hl Secondary transcription factors are turned on (or off) at
S11 various specific concentrations of those primary, gradient-
S12 fl forming factors. The secondary transcription factors then
fl Snout Snout interact with one another to sharpen the boundaries be-
tween their expression domains. They also modulate the
lj lj expression of yet more genes. Some of these downstream
genes themselves code for diffusible transcription factors
that set up their own concentration gradients. Thus, as the
neural tube develops and grows, an ever-increasing number
of crisscrossing gradients cause an increasing number of
other genes to be expressed in spatially restricted domains.
Thus, the neural tube gradually becomes divided into ever
more distinct territories.
BRAIN EXERCISE
How does rostrocaudal patterning in the midbrain and
forebrain differ from rostrocaudal patterning in the
hindbrain and spinal cord?
4.3 Where Do Neurons
ComeFrom?
Figure 4.13 Duplicating a cortical map.In some mouse embryos,
Fgf8was artificially expressed caudally as well as rostrally (compare At the early stages of embryonic development, the walls
toFigure 4.12 A). These embryos developed a second primary somato- of the neural tube consist almost exclusively of rapidly di-
sensory cortex (S12) just caudal to the normal primary somatosensory viding, undifferentiated cells that are called progenitors.
cortex (S11). Shown here is a section through the two somatosensory
areas, stained to reveal dark patches that represent distinct body As these cells divide, they make more progenitors just like
parts(hl = hindlimb, fl = forelimb, lj = lower jaw). Note that the two themselves. However, the progenitors divide more often
s omatosensory cortices are mirror images of one another. [Image from in some regions than in others, partly because of spatial
Assimacopoulos et al., 2012] variations in the transcription factors we just discussed.
Combined with fluid pressure from the ventricle inside the tube, the differential pro-
liferation of progenitors causes the neural tube to bulge in several places. Despite this
bulging, the walls of the neural tube retain a uniform thickness (Figure 4.14). That is,
the walls do not become thinner as a balloon does when you blow into it; nor do they
thicken as the progenitors proliferate. Instead, the walls of the young neural tube
expand only in surface area. They expand tangentially rather than radially.
Neurogenesis
The progenitor cells in the walls of the young neural tube are called radial cells (or radial
glia) because they have a process that extends radially away from the cell body toward
the brain’s external surface. Each progenitor’s cell body moves up and down within
this radial process. When the cell body is near the external brain surface, the progenitor
duplicates its chromosomal DNA (the cell is in S-phase). When the cell body sinks down
to the ventricular surface, where brain tissue borders the ventricle, the cell divides by
mitosis (Figure 4.14). After each cell division, the cell bodies of the two daughter cells
move radially away from the ventricular surface and then repeat the cycle. At some point,
however, the daughter cells do not reenter the cell cycle. Instead, one or both of them
stop dividing and begin to differentiate into neurons (Figure 4.15 A).
This process of ceasing to divide and then becoming a neuron is called neuro-
genesis. The time when a daughter cell leaves the proliferative cell cycle is called its
Where Do Neurons ComeFrom? 109
Tangential A C
Ventricular Radial MZ
zone
Forebrain Mitotic cells Newborn neurons
ventricle Midbrain VZ Progenitors
ventricle
(radial cells)
Ventricular surface
B
MZ Newborn neurons
and non-radial glia
Brain tissue 100 µm
SVZ Intermediate
Figure 4.14 The walls of early embryonic brains are thin.Shown progenitor
here is a horizontal section through the head of a young chicken VZ Progenitors
embryo (rostral is to the left) stained so that brain cells are blue. As (radial cells)
you can see, the brain at this stage consists of a relatively thin layer of
Ventricular surface
cells, called the ventricular zone. This ventricular zone surrounds large
fluid-filled ventricles. The illustrated section was also stained with an
antibody that stains cells in mitosis brown. The insert at the top right
shows that the mitotic nuclei are usually close to the ventricular sur- Figure 4.15 Neurogenesis and migration in the neocortex.Panel (A)
face. The black arrows indicate the radial and tangential dimensions in shows two progenitors with their cell bodies in the ventricular zone (VZ)
the ventricular zone. and a long, thin process extending radially through the mantle zone
birth date. As you will learn in Chapter 5, a few neurons (MZ). When a progenitor divides (red arrows), two daughter cells result.
One daughter migrates out of the VZ and becomes a newly born
and many glial cells are born after the organism itself is neuron. The other daughter remains a progenitor, which then divides
born, and some neurogenesis persists well into adulthood. into two newborn neurons. As shown in (B), some cells remain prolifera-
For now, let us focus on embryonic development, when tive after they leave the VZ and accumulate in the subventricular zone
the vast majority of neurons is produced. At the beginning (SVZ). When these “intermediate progenitors” divide, they generate
of embryonic neurogenesis, only one of the daughter cells neurons as well as astrocytes and oligodendrocytes. Panel (C) shows
young neurons migrating radially away from the ventricular surface
exits the cell cycle; the other daughter cell remains a pro- (bottom) along the slender processes of a radial cell. The newborn cells
genitor. Toward the end of embryonic neurogenesis, each in this image are progeny of a single progenitor that was infected with a
progenitor tends to produce two newborn cells. As you recombinant retrovirus. [A and B after Dehay and Kennedy, 2007; C from
would predict, this leads to a rapid decrease in the number Noctor et al., 2001]
of progenitors.
Many of the transcription factors that partition the neural tube into its various
subdivisions also affect neurogenesis timing. However, most transcription factors affect
neurogenesis indirectly by acting on a few key molecules. One of these crucial mole-
cules is beta-catenin (β-catenin; yes, the same molecule mentioned in Figure 4.1).
Within proliferating brain cells, β-catenin tends to delay neurogenesis onset. This
was demonstrated in experiments with transgenic mice in which β-catenin levels
were artificially elevated. The progenitors in these manipulated brains divided more
often than normal and therefore caused brains to grow much larger than normal.
Another molecule that plays a major role in controlling neurogenesis is delta-1. At
high levels of delta-1 expression, cells leave the cell cycle and begin neurogenesis. In
a sense, this molecule has an effect opposite to that of β-catenin.
Radial Neuronal Migration
As shown in Figure 4.16, newborn cells tend to migrate radially away from their place
of birth in the so-called ventricular zone and form a separate layer called the mantle
zone that extends from the ventricular zone to the external brain surface. At first the
mantle zone is thin, but it soon thickens enormously.
110 Chapter 4 Developing a Nervous System
CP Although the mantle zone contains mainly postprolif-
erative cells, there is one big exception to this rule. In
SP several regions of the telencephalon, some cells leave the
ventricular zone while they are still progenitors. These in-
100 µm SVZo termediate progenitors accumulate just outside of the
ventricular zone in what is called the subventricular zone.
MZ 55 65 72 SVZi Within the subventricular zone, the intermediate progeni-
VZ Embryonic age (days) 94 tors divide at least one more time (Figure 4.15 B). Eventually,
they give birth to young neurons and glial cells. An inter-
45 esting aspect of the subventricular zone is that it is much
thicker in the neocortex of primates than of non-primates.
Figure 4.16 Brain tissue thickens as development proceeds.Shown Its enlargement probably explains, at least in part, why the Neurons born (%)
here are drawings of radially oriented slivers through the neocortex neocortex is significantly larger in primates than in most
ofa monkey between 45 and 94 days of embryogenesis. Initially the other mammals, even after you account for differences in
neocortex consists almost entirely of proliferating cells that form the overall brain size.
ventricular zone (VZ). As development proceeds, cells migrate out of
the VZ to form more superficial layers, collectively referred to as the After newborn cells leave the ventricular and subven-
mantle zone (MZ). As the MZ thickens, it subdivides into multiple tricular zones, they tend to keep migrating radially. During
layers, including the internal and outer subventricular zones (SVZi this migration, the young cells maintain contact with the
andSVZo, respectively), the subplate (SP), and the cortical plate (CP). long processes of the radial cells, which extend all the way
[From Dehay and Kennedy, 2007] to the brain surface even as the mantle zone thickens
(Figure 4.15 C). In fact, newborn cells tend to use the radial
processes as a sort of guide rail or climbing rope.
Neurogenesis Timing and Cell Fate
In the neocortex, migrating young neurons burrow their
wAayNtehorcoourtgehx the older cells above them. They do not leave
their “radial monorail” until they approach the external
brain suI rface, where they encounter a molecular “please
disembIaI/rIIkI ” signal called reelin. Because each generation
of neocortical cells leaves the radial monorail after it has
passed tIVhrough most of the older cellVs, theIVadult neocortex
is structured so that the firstV-Iborn cells make uIpI/IIIthe deep
corticalVlayers, whereas the younger cells occupy progres-
sively mVoI re superficial layers (Figure 4.17 A). This orderly
arrangement is u3n5usual, h4o5wever. I6n0 most r8e5gions o1f2t0he
Vnenetrrivcuolaurs system, youTnigmcee(ldlsadysoonfoetmcobnrysoisgteenntelysism) igrate
surface
A Neocortex B Retina
I Neurons born (%) BP Neurons born (%) R
II/III RGC C
V IV
IV VI II/III RGC
V BP
VI
R&C
Ventricular
surface 35 45 60 85 120 Ventricular 35 40 45 50 60 70 90
surface Time (days of embryogenesis)
Time (days of embryogenesis)
B Retina
Figure 4.17 Neurogenesis timing in neocortex and retina.Illustrated in (A) are a stained section through the neocortex of an adult monkey (left;
Neurons born (%)
Rfvueorlmtohpaemnr fnernuotmm.RTeGthrhCaieslsovierndndetirrcilacytureleadlraisstRiutGoirnnCfcasthcleiap.yCiesornsn)oeatpnshdeCoeatnogirrneaBcpthePhpestrhoeortsiwn(Cian)(gBte)t.hnInadtstRtnoeeabuder,obrneostrinninabtlehgfeaonuregplrpiooednr cortical layers are born at progressively later times in de-
cells (RGC) are born before bipolar cells (BP) and end up
photoreceptors (R), but both cell types are found in the
same layer oBfPthe adult retina. Although the correlation between spatial location and birth date is less orderly in the retina than in the neocortex,
time of birth in both regions plays at least some role in determining the neuron’s fate. [A courtesy of Pasko Rakic, MD, PhD; B after Rapaport et al.,
1996, and image from Anger et al., 2004]
R&C
Ventricular 35 40 45 50 60 70 90
surface Time (days of embryogenesis)
How Do Axons Find Their Targets? 111
past older cells. Instead, they tend to get off the radial monorail just before they
reach the older cells. In the retina, another region in which neurogenesis has been
studied extensively, some young neurons migrate past older cells, but others do not
(Figure 4.17 B). Thus, different regions exhibit different patterns of radial neuronal
migration.
The time at which a cell is born affects not only its position in the adult brain
but also the type of cell it will become (its fate). Progenitors generally produce
neurons before they produce glial cells, and they tend to make oligodendrocytes
(the glia that insulate axons with myelin; see Chapter 2) before they make astrocytes
(see Chapter 5). Moreover, progenitors give birth to different types of neurons at
different times (Figure 4.17). Even in tissue culture, young progenitors typically pro-
duce different types of cells than old progenitors. This suggests that the progenitors
themselves are changing over time. Consistent with this hypothesis, the mix of tran-
scription factors expressed by various progenitors varies with age.
Studying Cell Type Specification with Transplantation Experiments
Important insights into cell type specification were obtained by Sue McConnell and
her collaborators. They transplanted young neocortical progenitors, which normally
generate cells of the deep cortical layers (see Fig 4.17 A), into the neocortex of older
embryos and found that the offspring of the transplanted cells ended up in the upper
cortical layers. The transplanted young progenitors acted like old progenitors. In
contrast, when McConnell transplanted old progenitors into younger embryos, the
transplanted progenitors acted just as old progenitors normally do: they generated
neurons in the upper cortical layers. These experiments revealed that young progeni-
tors are competent to generate a wide variety of different cell types; with age, their
competence narrows. This probably means that some of the molecular changes oc-
curring inside of the progenitors are irreversible. A major unresolved question is
what causes these fateful changes in gene expression. Some evidence suggests that
young neocortical neurons send signals back to the progenitors, instructing them to
start making a different type of cell. This would explain why young progenitors trans-
planted into the older neocortex act like the old progenitors.
Recapping our discussion thus far, a cell’s fate depends primarily on when and
where it was born. In a way, young cells are like young people. Their behavior de-
pends, at least in part, on when and where they were born. The analogy is also apt
because, like humans, many neurons stay close to where they were born. They usually
migrate along radial processes, but most of these migrations are relatively short. Of
course, there are exceptions to this rule. For example, many young neurons in ro-
dents migrate from the ventral telencephalon into the olfactory bulb and into the
neocortex, where they become GABAergic interneurons. The human neocortex
seems to harbor fewer of these long-distance “immigrants” than the rodent neocor-
tex does, but human brains do exhibit some long tangential migrations. Specifi-
cally, in humans some young cells migrate tangentially out of the telencephalon into
the dorsal thalamus. This seems to be a distinctly human trait. In any case, such long
tangential migrations are rare. Most young brain cells migrate radially and do not
wander far.
BRAIN EXERCISE
Given the data shown in Figure 4.17 B, which types of retinal cells must be
migrating past which other types of retinal cells during development?
4.4 How Do Axons Find Their Targets?
So far we have discussed how the nervous systems develop a variety of distinct cell
types and how those cells arrive at their proper locations. By and large, these cells do
not yet have complete axons. How, then, do the connections between neurons form?
This is a difficult problem because each neuron’s axon must grow toward, and ulti-
mately synapse with, a very specific subset of neurons. It must find a few needles in an
112 Chapter 4 Developing a Nervous System
Figure 4.18 Growth cones.Panel (A) is a A B C Actin
drawing by Ramón y Cajal showing axons Sensory Filopodia filaments
of dorsal sensory neurons in the spinal neurons
cord sending their axons ventrally. Each Microtubules
growing axon is tipped with a growth Growth cones
cone, and some of the axons have crossed
the ventral midline. Panel (B) shows that
growth cones can have a variety of shapes,
depending mainly on the kind of space
they are growing through. Panel (C) shows
actin filaments in red and microtubules in
green. [A from Cajal, 1890; B after Bray,
1982; C courtesy of Paul Letourneau, PhD]
enormous proverbial haystack! To understand how neurons accomplish this task,
you must first learn how axons grow.
Axonal Growth Cones
Axons typically grow out of a neuron’s cell body shortly before the neuron ends its
migration. Growing axons are filled with long and slender microtubules and tipped
with growth cones (Figure 4.18). Ramón y Cajal originally described these growth
cones as “a sort of club or battering ram, endowed with an exquisite chemical sensi-
tivity, with rapid amoeboid movements, and with impulsive force by which it is able
to proceed forward and overcome obstacles met in the way, forcing cellular inter-
stices, until it arrives at its destination” (Ramon y Cajal, 1917, p. 599). This descrip-
tion is remarkably accurate, considering that Cajal based it entirely on what he saw in
fixed and stained tissue sections. Modern scientists can watch time-lapse movies of
growth cones wandering around a tissue culture dish or in a living brain slice, but
their observations have generally confirmed Cajal’s account.
Extending and Retracting Filopodia
Modern science also revealed that growth cones regularly extend and then retract
slender protrusions called filopodia (Figure 4.18). These filopodia are filled with
actin filaments that are linked to microtubules near the growth cone’s center. The
actin filaments are also anchored to the cell membrane near the filopodial tips. Indi-
vidual actin subunits are regularly added to the end of the actin filament that ex-
tends into the filopodia. This tends to push the filopodia forward, lengthening
them. However, even as the actin filaments extend, they are pulled back toward the
growth cone’s center by myosin molecules (you will learn about actin and myosin in
Chapter 8). When the rate at which the actin filaments are pulled backward exceeds
the rate at which actin is added at the tip, the filopodia retract. This sounds straight-
forward, but you might wonder: what happens to the basal ends of the actin filaments
as they are pulled backward? Do they pile up inside the growth cone? No, they do not.
Instead, actin subunits are regularly removed from the base of the actin filaments.
Filopodia can make a growth cone move, but only if the filopodia tips adhere to
something in the external environment. Think of it like this: if a filopodium is re-
tracted while its tip is stuck on some other cell or extracellular material, then the tip
cannot move back toward the growth cone. Instead, the growth cone must move
toward the filopodium tip. It is like pulling on a rope that is attached to the ceiling;
when you pull on the rope, you end up pulling yourself off the floor. The force exerted
by filopodial retraction can be visualized when axons are grown in tissue culture.
Sometimes, when a filopodium from a cultured growth cone contacts the process of
another neuron, it pulls on the other process (Figure 4.19). If you imagine this other
process as an immovable extracellular object, then you can visualize the growth cone
being pulled forward toward the filopodial tip. As the growth cone moves forward,
the axon behind it elongates.
How Do Axons Find Their Targets? 113
AB Figure 4.19 Growth cones can generate
traction.Shown here are two frames from
A passing axon Growth cone a time-lapse video of axons growing in
10 µm culture. Some of the filopodia that have
adhered to a passing axon in (A) have
r etracted by frame (B) and are now pulling
the axon toward the growth cone’s center.
This demonstrates that retracting filopo-
dia may exert considerable force. The
a sterisks in A and B indicate a speck of
dirtthat remains in a constant position
between the two video frames. [From
Heidemann et al., 1990]
Changing an Axon’s Direction of Growth
Once you understand how axons elongate, you can grasp how they turn. Other things
being equal, a growing axon will turn toward whatever locations offer the best “foot-
holds” for its filopodia. This is only true, however, if the filopodia extend equally in
all directions. If they extend mainly toward the left, for example, then the growth
cone will probably turn left, even if the environment is slightly stickier on the right
side. Thus, a growth cone’s direction of movement is a product of (a) how well its
filopodia stick to their surroundings and (b) how likely the growth cone is to extend
its filopodia in a particular direction. This explains why growth cone guidance factors
generally affect either the strength with which filopodia stick to their substrates or
the likelihood of filopodia extending in a specific direction, or both.
Growth Cone Guidance
Ramón y Cajal proposed that growth cones “become oriented by chemical stimula-
tion, and move toward the secreted products of certain cells” (Cajal, 1894, p. 146).
This idea was appealing because many microorganisms exhibit such chemotaxis
(movement toward a chemical). However, Cajal had no direct evidence for his
hypothesis.
Evidence for Growth Cone Chemotaxis
The first convincing evidence for growth cone chemotaxis was obtained in the 1990s
when researchers filled glass micropipettes with hypothesized guidance factors and
then positioned the pipettes next to the growth cones of axons growing in tissue cul-
ture (Figure 4.20). As the molecules diffused out of the pipettes (or were pushed out
slowly), a concentration gradient formed in the culture dish. Within about 30 minutes,
Netrin - 0 min Netrin - 60 min Intact netrin
Micropipette Growth cone Figure 4.20 Axon chemotaxis.Shown in
the top row are photographs of an axon
Axon growing in culture. After 60 minutes the
axon has grown toward the tip of a glass
20 µm micropipette filled with the molecule
netrin. What you cannot see in these
Heated netrin - 0 min Heated netrin - 60 min Heated netrin photographs is that the experimenters
areslowly pushing the netrin out of the
pipette, setting up a netrin concentration
gradient within the culture dish. The
graph on the right shows growth trajecto-
ries for 10 different axons. All of them
grew toward the netrin-filled pipette. The
bottom row illustrates the results from a
control experiment in which the netrin
was inactivated by heating it to 75°C.
[From Ming et al. 1997]
114 Chapter 4 Developing a Nervous System
the growth cones moved up this concentration gradient toward the pipette tip. This
finding strongly supported Cajal’s chemotaxis hypothesis.
Since those early studies, scientists have learned that growth cones can detect
surprisingly small changes in molecule concentrations. Apparently, growth cones do
this by moving slowly (<1 µm/minute), sampling their chemical environment repeat-
edly, and then integrating this information over time. The process is analogous to
how you might locate a delicious meal by wandering around, sniffing repeatedly, and
moving up the concentration gradient of odors given off by hot pizza. Scientists have
also discovered that growth cones are repelled, rather than attracted, by some mole-
cules. Proteins of the semaphorin family are particularly effective growth cone re-
pellents. When growth cones contact such repellent molecules, the filopodia tend to
collapse on the side with the highest concentration of the offending molecule. This
differential collapse of filopodia forces the growth cone to turn away from the source
of the repulsive molecules.
One problem for Cajal’s chemotaxis hypothesis is that gradients of diffusible mol-
ecules tend to be steep, which means that growth cones can be guided by diffusion-
based concentration gradients only over distances of less than a few hundred
microns. Axons solve this problem in two ways. First, they use diffusion gradients
mainly at early stages of development, when embryos are so small that a single
concentration gradient spans a significant fraction of the embryo’s nervous system.
Second, growing axons use different guidance cues for different segments of their
journey, much as driving directions are usually broken down into multiple segments
(go North on the main road until you get to the stop sign, then turn right and go up
the hill, etc.). This is a clever trick. However, journey segmentation creates a new
problem: if an axon finds the first target in its journey by growing up a concentration
gradient centered on the first target, how can it leave the first target and move on to
the second one? It would seem that the growing axon would have to be reprogrammed
when it reaches the first target. It must stop being attracted to the molecules secreted
by the first target and start responding to new cues in novel ways.
Axon Reprogramming
Strong evidence for axon reprogramming has been obtained from research on com-
missural neurons in the spinal cord, which convey information from one side of the
spinal cord to the other. The axons of these commissural neurons course ventrally
toward the floor plate, cross the midline, and then ascend again on the other side of
the spinal cord (Figure 4.21). In 1988, Marc Tessier-Lavigne and his collaborators
discovered that the axons of commissural neurons grow ventrally because they are
attracted to a diffusible substance secreted by floor plate cells. Six years later, this
substance was identified as netrin (from the Sanskrit word for “he who guides”). Why
do the commissural axons exit from the floor plate after crossing the midline? Be-
cause when commissural axons get to the floor plate, they start making a membrane
receptor called robo (short for roundabout, which describes the abnormal growth of
the commissural axons when this receptor is knocked out). The robo receptors make
Figure 4.21 Netrin and slit are expressed A B
in the floor plate.Shown in (A) is a section D Midbrain
through an embryonic spinal cord of a V
chick, stained to reveal netrin expression Commissural axon
in the floor plate. Shown in yellow is a
commissural (midline crossing) neuron Telencephalon Slit
whose axon grows ventrally toward the Medulla
netrin source, crosses to the other side, Netrin Spinal cord
and then grows away from the floor plate.
Shown in (B) is the expression pattern of
slit, another molecule that is produced in
the floor plate; the embryo shown here is
from a mouse. [A from Kennedy et al.,
1994; B from Holmes et al., 1998]
How Do Axons Find Their Targets? 115
Slit Netrin Slit Netrin
Robo DCC Robo DCC
Repulsion Attraction Repulsion Attraction
Netrin and Slit
Figure 4.22 Axon reprogramming.The axons of commissural neurons initially grow toward the
floor plate because they are attracted by netrin (left). This positive chemotaxis involves netrin
a ctivating its receptor Deleted in Colorectal Cancer (DCC) on the commissural axons. Once the
axons have crossed the floor plate, they express high levels of robo, which is activated by slit.
Roborepresses the DCC receptor, making the growth cone less attracted to netrin. In addition,
robocauses the growth cones to be repulsed by slit (right). As a result of all these molecular inter
actions, the axons are first attracted to the floor plate, and then repulsed by it. Arrows in the two
flowcharts indicate p ositive interactions; T-shaped lines represent inhibition. Bold type indicates
ahigh concentration of the specified molecule.
the growth cone less sensitive to netrin and, simultaneously, cause it to be repulsed
by a molecule called slit, which is also secreted by floor plate cells. Because of these
changes, growth cones that were initially attracted by netrin become indifferent to
netrin and repulsed by slit, which they had hitherto ignored (Figure 4.22). Thus, the
axons get reprogrammed from one step of their journey to the next.
Substrate-bound Axon Guidance Molecules
In the years since netrin was identified, neuroscientists have discovered many addi-
tional diffusible axon guidance molecules. They have also discovered guidance
factors that do not diffuse freely but are, instead, attached to cell membranes or extra
cellular material. Many of these substrate-bound factors attract axons by offering the
growth cones ideal levels of “stickiness” for growth; others promote filopodial exten-
sion. Some substrate-bound factors encourage growth cones to crawl along the surface
of other axons that express the growth-promoting molecules on their surface. This
is a good idea because it allows late-born neurons to find their targets by simply fol-
lowing the axons of neurons that were born earlier and have already blazed a trail.
Because the trailblazing pioneer axons are growing out when the embryo is very
small, they can navigate by the range-limited diffusion gradients. Later neurons cannot
use the diffusion gradients, but they can follow the pioneers to the appropriate target.
This helps explain why many axons travel in large bundles called nerves or axon tracts.
Of course, follower axons must at some point leave the other axons and find their own,
unique targets. This process of defasciculation (fasciculus means “bundle”) also requires
growth cone reprogramming, but has not yet been studied in detail.
The Retinotectal System
Extensive work on the mechanisms of axon guidance has been conducted on non-
mammalian vertebrates, specifically on the projections from the retina to the mid-
brain’s optic tectum, which is hom*ologous to the mammalian superior colliculus
and involved in orienting the eyes and head toward visual targets (see Chapter 11).
The retinotectal system is well suited to research on axon guidance because both
the retina and the optic tectum are relatively large and flat (sheet-like) in structure.
Moreover, the retinotectal projections exhibit precise topography, with nasal retina
projecting to the caudal tectum, temporal retina projecting to the rostral tectum,
116 Chapter 4 Developing a Nervous System
ventral retina projecting to medial tectum, and dorsal retina
projecting to the lateral tectum (Figure 4.23).
Medial
Classic experiments on the retinotectal system were per-
formed by Roger Sperry in the 1940s. Particularly influential
Rostral Tectum Caudal was a study in which Sperry crushed the optic nerve of
an adult salamander and surgically rotated the affected eye
by 180°. The crushed axons eventually grew back into the
Lateral tectum, but the experimental animals now acted as if their
visual world was inverted (on the side viewed by the rotated
eye). When food was presented in front of the animal, it ori-
ented toward its tail; and when food was presented above the
salamander, the animal looked down. These behavioral ob-
servations led Sperry to conclude that the regenerating axons
must have grown back to their original targets in the optic
tectum but then provided incorrect, inverted information
Dorsal
about the location of visual stimuli.
Based on this inference, Sperry developed his chemo
Retina Nasal Temporal
affinity hypothesis. Sperry proposed that each retinal axon
Ventral expresses a distinct set of molecular markers, and the tectal
neurons express matching (complementary) markers. As the
retinal axons grow into the tectum (both in embryonic devel-
opment and during regeneration after injury, as in Sperry’s
Figure 4.23 The retinotectal system.The projections from the experiment), they seek out and selectively terminate on tectal
retina to the optic tectum (the superior colliculus in mammals) are
crossed and topographic. Nasal retina projects to caudal tectum, neurons with the matching marker. In essence, each growing
temporal retina to rostral tectum, dorsal retina to lateral tectum, axon has a “chemical affinity” for a specific target neuron. Be-
andventral retina to medial tectum. cause it is improbable that every neuron expresses a unique
chemical tag, Sperry proposed that the hypothesized markers are expressed in two or
more intersecting gradients across the retina as well as the tectum (color gradients in
Figure 4.23). Although Sperry’s hypothesis was prescient, it took many years before
neuroscientists discovered which molecules were expressed in the hypothesized gra-
dients and how those molecules control axon guidance.
Crucial progress in the search for Sperry’s hypothesized molecular tags came from in
vitro experiments that used a so-called stripe assay. The first step in these experiments
was to deposit membranes from cells in the rostral and caudal parts of the tectum onto a
piece of filter paper so that membranes from rostral and caudal tectum formed alternating
stripes (Figure 4.24). The experimenters then took a horizontal strip of embryonic retina,
placed it right next to the alternating stripes in a tissue culture dish, and watched the reti-
nal axons grow onto the tectal membrane stripes. Retinal neurons that normally sit close
to the nose (in the nasal retina) grew indiscriminately onto both sorts of stripes. However,
neurons that normally sit toward the side of the head (in the temporal retina) preferred
growing onto stripes of membranes from the rostral tectum. When the tectal membranes
were treated to remove or destroy most proteins, the temporal retinal axons lost their pref-
erence for rostral tectum. This showed that cells in the caudal tectum contain a mem-
brane-bound protein that repels temporal, but not nasal, retinal axons.
The Role of Ephrins in Axon Guidance
The repulsive factor in the caudal tectum turned out to be a molecule of the ephrin
family that is expressed in a smooth gradient across the tectum, with the highest
ephrin levels in the most caudal tectum. It has also become apparent that axons from
the nasal retina express low levels of the matching ephrin receptor, whereas axons
from more temporal regions of the retina express progressively more ephrin receptor.
This difference in ephrin receptor expression makes axons from the nasal retina less
sensitive to ephrin’s repulsive effects. Add to this the finding that ephrin expression
is highest in the caudal tectum, and you come up with the following simple model: as
retinal axons grow from the tectum’s rostral edge toward the back, they simply grow
until the ephrin-induced repulsion becomes too much for them to bear. Because the
How Do Axons Find Their Targets? 117
A Nasal retina B Temporal retina A Normal retina B Mutant retina
Caudal tectum Caudal tectum
Rostral tectum Rostral tectum
Explanted cell bodies
Explanted cell bodies Superior colliculus Superior colliculus
Ephrin-A coated Uncoated Nasal Temporal Nasal Temporal
Normal retina Mutant retina
C
Growing axons Rostral Rostral
Cell bodies of explanted retinal neurons
Nasal Temporal
Figure 4.24 Ephrin guides retinal axons.Cultured neurons from nasal Superior
or temporal retina were prodded to extend their axons on top of cell colliculus
membranes from either rostral or caudal tectum that had been deposited
Caudal Caudal
in alternating stripes on a piece of filter paper. The nasal retinal axons
grew indiscriminately on both sets of membranes, but the temporal
axons grew only on membranes from rostral tectum. A membrane-
bound protein called ephrin helps to guide the retinal axons. Panel (C)
shows what happens when nasal and temporal axons are prodded to Figure 4.25 Ephrin gradients control map formation.In normal
grow on stripes that are coated with ephrin-A (faintly reddish lanes) or mice (A), nasal retina projects only to caudal tectum. This is not the
not (darker lanes). As you can see, the temporal axons avoid the ephrin case in transgenic mice (B) engineered so that a subset of retinal
molecules. [A and B from Walter et al., 1987; C from Monschau et al., 1997] neurons (indicated by green rings) express more than their normal
amount of ephrin receptor (concentration indicated by intensity of
nasal axons are less sensitive to ephrin than the temporal reddish-brown shading). The axons of these transgenic neurons
axons, they are more likely to end up growing all the way target collicular cells that express very low levels of ephrin (light blue
into the caudal tectum. Because both ephrin expression in shading). Therefore, the retinal cells in the transgenic mice form two
separate collicular maps. One map is formed by the normal retinal
the tectum and ephrin receptor expression in the retina neurons, the other one by the transgenic cells. Consistent with this
follow smooth gradients, this model predicts the smooth hypothesis, the nasal retina in the transgenic animals projects to
rostrocaudal topography that the retinotectal projection twoseparate locations in the superior colliculus. The illustrated
normally exhibits. retinalooks like a cross because it was cut before it was flattened.
[After Brown et al., 2000]
To test this ephrin gradient hypothesis, researchers cre-
ated transgenic mice that overexpress ephrin receptors in a randomly scattered subset of
retinal neurons (Figure 4.25). These mice contain two superimposed gradients of retinal
ephrin receptor expression: one normal gradient and one comprising the mutant cells. Im-
portantly, when the scientists examined the retinal projections in these transgenic mice,
they found two separate topographic sets of projections to the superior colliculus (the ho-
molog of the optic tectum). Axons from the normal retinal neurons form a map in the
caudal half of the superior colliculus, and axons from the mutant neurons form a second
map in the rostral superior colliculus. This finding confirms that retinal axons expressing
high levels of ephrin receptor abhor the caudal superior colliculus. The data also showed
that a retinal axon’s target is not determined by the absolute amount of ephrin receptor it
expresses. If it were, then some of the mutant retinal neurons should have found no targets
in the superior colliculus at all (because absolute ephrin receptor levels are higher in some
of the mutant retinal cells than they are in any normal retinal neurons). Instead, a retinal
axon’s target in the superior colliculus is determined by the amount of ephrin receptor the
axon expresses relative to the amount of ephrin receptor present in the other retinal axons.
Essentially, the retinal axons are all jockeying for target space in the superior colliculus.
118 Chapter 4 Developing a Nervous System
Eventually each retinal axon ends up with as good a target as it can get, given the competi-
tion. Overall, these data are consistent with Sperry’s chemoaffinity hypothesis, but they
also reveal that an axon’s target is more flexible—more dependent on competition with
other axons—than Sperry had originally envisioned.
Recent studies have filled in further details about the formation of topographic maps.
They have shown, for example, that ephrins may be attractive as well as repulsive, and
that a different member of the ephrin family is involved in establishing how retinal
axons map onto the mediolateral (rather than rostrocaudal) axis of the superior collicu-
lus. In fact, graded patterns of ephrin expression are seen all across the nervous system
and appear to be crucial for the formation of many different topographic projections. This
plethora of ephrin gradients is interesting because it suggests that substrate-bound mole-
cules like ephrin are a powerful means of developing orderly axonal projections. The idea
of diffusible signals such as netrin or semaphorin guiding axons at a distance is intuitively
appealing, but such signals are not required. As long as the substrate-bound molecules are
expressed in gradients, growing axons can use them to find their targets. The axons simply
grow up or down such gradients until they find their “sweet spot” in the gradient. Once
there, the axons slow their forward growth, branch repeatedly to form terminal arboriza-
tions, and begin to make some synapses.
BRAIN EXERCISE
When experimenters transplant embryonic neurons into adult nervous systems
in the hope of repairing brain damage, the axons of the transplanted neurons
usually fail to reach their normal targets. Why is that?
4.5 How Do Synapses Form?
Once an axon has found its general target, it needs to form synapses at specific locations
within the target area. Because most presynaptic terminals are directly opposed to
postsynaptic sites, the mechanisms of synapse formation must somehow assure that
presynaptic terminals are located directly across from postsyn-
5 min aptic specializations. In addition, glutamatergic terminals must
A B T=0 be matched up with postsynaptic sites that contain glutamate
Dendrite
Filopodia 6 min receptors; GABAergic terminals must be matched up with post-
1 min synaptic GABA receptors; and so forth for each kind of neu-
rotransmitter. Finally, some axons terminate specifically on the
cell bodies of their target neurons, whereas others prefer to end
on their dendrites. How does such specificity arise?
2 min 7 min Perhaps postsynaptic specializations exist before the pre-
synaptic terminals arrive, and the presynaptic axons simply
sniff them out by following some molecular trail. This idea is
Cell body 3 min 8 min appealing but false. Instead, synapse formation begins with
pre- and postsynaptic cells extending filopodial “feelers” that
contact each other. Once contact has been established, the pre-
4 min 9 min and postsynaptic processes communicate. If the exchanged
signals are positive, then pre- and postsynaptic specializations
begin to form. Over time, these molecular interactions
strengthen the bond between the pre- and postsynaptic cells, and
a mature synapse is formed. Thus, postsynaptic specializations
do not exist before the axon terminals arrive. Instead, they are
Figure 4.26 Time-lapse video reveals dendritic filopodia move- laid down as synapses emerge. Now that you have the basic
ments.The large image on the left (A) depicts a neuron from the neo- idea, let us get into details.
cortex of a 2-day-old mouse injected with a fluorescent dye so that its
processes could be visualized in the living animal. The area in the small Filopodial Interactions
white rectangle is shown at higher magnification in (B), which shows 10
frames from a time-lapse video taken at one-minute intervals. Compar- When young dendrites first grow out of the cell body, they are
ing successive frames, you can see dendritic filopodia extending from tipped with growth cones very similar to those of growing
the growing dendrite and then retracting. Scale bars equal 25 µm in A axons. Soon after the initial outgrowth, however, young
and 5 µm in B. [From Portera-Cailliau et al., 2003]
How Do Synapses Form? 119
dendrites distinguish themselves from young axons by Wild-type fly Furry mutant
growing thicker. Time-lapse movies further reveal that
dendrites send out long and slender filopodia all along Figure 4.27 Dendritic tiling.Shown on the left are two sensory
their length, not just at the tip (Figure 4.26). These den- n eurons in a normal, wild-type (wt) fruit fly. As you can see, the two
dritic filopodia either retract soon after forming, or they dendrites don’t overlap, a phenomenon referred to as dendritic tiling.
develop into new dendritic branches. Some such behind- Shown on the right are two neurons of the same type in a mutant fly
the-tip filopodia are also seen in axons, particularly as they with a defective furry (Fry) gene. In such mutants, dendritic tiling is
approach their target and form terminal arborizations, but disrupted. Even within a single dendritic tree, branches cross abnor-
they are more frequent in dendrites. Dendritic filopodia mally often (arrowheads). [After Emoto et al., 2004]
probably sample their chemical environment just as axonal
filopodia do. Good evidence for this hypothesis comes
from the observation that dendrites from different cells of
the same cell type tend to exhibit little or no overlap
(Figure 4.27). Most likely, this dendritic tiling arises be-
cause growing dendrites sense nearby dendrites of the
same type and are repelled by them. This tendency to be
repulsed by dendrites of the same type also ensures that
the dendritic branches of a single neuron are spread far
and wide rather than clumped. This spreading of the den-
dritic tree, which is very apparent in cerebellar Purkinje
cells (Figure 4.28), maximizes a neuron’s ability to gather
inputs from many different axons.
Axon-sensing Dendrites? Figure 4.28 Purkinje cells’ planar
If dendrites can sense other dendrites, can they also sense axons? This question dendritic trees. Shown in (A) is a cerebellar
remains unanswered. However, some axons do release smallAamounts of neu- Purkinje cell from an adult human. Its
rotransmitter before proper synaptic contacts have formed, and some dendritic spectacular dendritic tree is largely con-
filopodia contain neurotransmitter receptors. Moreover, researchers have found fined to tCheerpeblaenllearofPtuhrekipnjaepceerl(lor screen)
that blocking transmitter release during development reduces the number of syn- in front of yoiun. fTrhoenrteaflovriee,wPurkinje cells
apses eventually formed. These observations suggest, but hardly prove, that some look much skinnier in “profile,” as shown
sort of communication between axons and dendrites helps their respective filopo- in (B). A major source of input to the
dia find one another in the vast wilderness of brain space. Additional communica- Purkinje cells are cerebellar granule cells,
tion probably occurs after a dendritic filopodium has contacted an axonal whose axons ascend to the level of the
filopodium. This would explain why filopodial contacts between GABAergic Purkinje cells, branch in a T-like fashion,
axons and GABA receptor-containing dendrites are more stable than contacts be- and then course at right angles to the
tween axons and dendrites that are not so nicely matched. Apparently, axons and plane of the Purkinje cell dendrites. This
dendrites, having made contact, check each other out to see whether the axon’s orthogonal orientation, together with the
transmitter is a good match for the postsynaptic cell. If the two match, then syn- large span of the Purkinje cell dendrites,
apse formation proceeds. allows each Purkinje cell to gather input
from more than 100,000 granule cell
axons. [A after Burns, 1911]
A B Granule cell
axons
Cerebellar Purkinje cell
in frontal view
Purkinje cells in profile view
Cerebellar granule cells
B Granule cell
axons
