This excerpt from

The Computational Brain.

Patricia Churchland and Terrence J. Sejnowski.
© 1994 The MIT Press.

is provided in screen-viewable form for personal use only by members

of MIT CogNet.

Unauthorized use or dissemination of this information is expressly

forbidden.

If you have any questions about this material, please contact

cognetadmin@cognet.mit.edu.
 Neuroscience Overview
2

1 INTRODUCTION

If we are to understand how the brain sees, learns, and is aware, we must
understand the architecture of the brain itself . The brain ' s computational style
and the principles governing its function are not manifest to a casual inspection
. Nor can they be just inferred &om behavior , detailed though the behavioral
descriptions may be, for the behavior is compatible with a huge number
of very different computational hypotheses , only one of which may be true of
the brain . Moreover , trying to guess the governing principles by drawing on
existing engineering ideas has resulted in surprisingly little progress in understanding
the brain , and the unavoidable conclusion is that there is no substitute
for conjuring the ideas in the context of observations about real nervous systems
: &om the properties of neurons and the way neurons are interconnected .
This chapter focuses on the " neuroscience" component of the " computa -
"
tional neuroscience synergy . Ideally , computer modelers should know as
much neuroscience as practising neuroscientists . In fact , however , there is too
much neuroscience to be thoroughly mastered even by a single neuroscientist .
An anatomist may know a lot about his designated region of the visual cortex ,
rather less about other cortical areas and subcortical brain structures , less again
about central pattern generation in the spinal cord , and even less about plasticity
of the vestibulo -ocular reflex . Our aim is to prepare the reader, &om whatever
constituency , for the general conceptual &amework we deploy and the
specific neurobiological examples discussed within that &amework . Consequently
, the material in this chapter is organized to cohere with a computa -
tional approach to exploring certain aspects of nervous system function .
Because levels turn out to be pivotal in our grand scheme of things , charac-
terizing levels in neurobiology is a matter of the first importance . The presentation
in this chapter is therefore keyed to illustrating anatomical and phys -
iological properties seen at different levels. Although understanding the
techniques whereby neurobiological data are gathered is also essential, to keep
the wagons moving we elected to provide this in the appendix at the end. Although
this chapter is meant to provide some basic neuroscience background ,

Substantial
portionsof thischapteraretakenfromSejnowski
andChurch
land(1989).
in the context of specific neurocomputational models
. relevant neuroscience
will be introduced .

IN NERVOUS
1. LEVELS SYSTEMS

Discussions concerning the nature of psychological phenomena and their neu-

" "
robiological bases invariably make reference to the notion of levels. In trying
" "
to be a bit more precise about what is meant by leveL we found three
different ideas about levels in the literature : levelsof analysis, levelsof organization
, and levelsof processing.Roughly speaking, the distinctions are drawn along
the following lines: levels of organization are essentially anatomical , and refer
to a hierarchy of components and to structures comprising these components .
Levels of processing are physiological , and refer to the location of a process
relative to the transducers and muscles. Levels of analysis are conceptual ,
and refer to different kinds of questions asked about how the brain performs a
task: into what subtasks does the brain divide the tasks, what processing steps
execute a subtask, and what physical structures carry out the steps? In what
follows , we elaborate on these distinctions .

Levelsof Analysis

A framework for a theory of levels, articulated by Marr (1982 ), provided an

important and influential background for thinking about levels in the context
of computation by nervous structures . ! This framework drew upon the conception
of levels in computer science, and accordingly Marr characterized
three levels: ( 1) the computational level of abstract problem analysis, decomposing
the task (e.g ., determining the 3- D depth of objects from the 2-D
pattern on the retina ) into its main constituents ; (2 ) the level of the algorithm ,
specifying a formal procedure to perform the task so that for a given input ,
the correct output results; and (3 ) the level of physical implementation , constructing
a working device using a particular technology . This division really
corresponds to three different sorts of questions that can be raised about
a phenomenon : ( 1) how does the problem decompose into parts?, (2) what
principles govern how the parts interact to solve the problem ?, and (3 ) what
is the stuff whose causal interactions implement the principles ?
'
An important element in Marr s view was that a higher - level question was
largely independent of the levels below it , and hence computational problems
of the highest level could be analyzed independently of understanding the
algorithm which performs the computation . Similarly , the algorithmic problem
of the second level was thought to be solvable independently of understanding
its physical implementation . Thus his preferred strategy was top -down
rather than bottom -up . At least this was the official doctrine though , in practice
'
, downward glances figured significantly in Marr s attempts to find problem
analyses and algorithmic solutions . Ironically , given his advocacy of the
- ' -
top down strategy , Marr s work was itself highly influenced by neurobiologi
cal considerations , and implementation facts constrained his choice of problem

Chapter2
 and nurtured his computatioriaI and algorithmic insights. Publicly, the advo-
'
cacy of the top-down strategy did carry the implication, dismaying for some
and comforting for others, that neurobiological facts could be more or less
ignored, sincethey were, after all, just at the implementationlevel.
Unfortunately, two very different issueswere confusedin the doctrine of
independence . One concernswhether, as a matterof discovery , one can figure
out the relevant algorithm and the problem analysis independently of facts
about implementation. The other concernswhether, asa matteroffonna' theory,
a given algorithm which is alreadyknown to perform a task in a given machine
(e.g ., the brain) can be implementedin someother machinewhich hasadifferent
architecture. So far as the latter is concerned, what computationaltheory
tells us is that an algorithm can be run on different machines,and in that sense
and that sensealone, the algorithm is independentof the implementation. The
formal point is straightforward: sincean algorithm is formal, no specificphysical
+
parameters(e.g ., vacuumtubes, Ca2 ) arepart of the algorithm.
That said, it is important to seethat the purely fonnal point cannot speakto
the issueof how best to discover the algorithm in fact used by a given machine
, nor how best to arrive at the neurobiologically adequatetask analysis.
Certainly it cannot tell us that the discovery of the algorithms relevant to
cognitive functions will be independent of a detailed understanding of the
nervous system. Moreover, it does not tell us that any implementation is as
good as any other. And it had better not, since different implementations
display enormousdifferencesin speed, size, efficiency, elegance, etc. The formal
independenceof algorithm from architectureis somethingwe can exploit
to build computationally equivalent machinesonce we know how the brain
works, but it is no guide to discovery if we do not know how the brain works.
The issue of independenceof levels marks a major conceptualdifference
betweenMarr (1982) and the current generationof researchersstudying neural
and connectionistmodels. In contrast to the doctrine of independence , current
researchsuggeststhat considerationsof implementationplaya vital role in the
kinds of algorithms that are devised and the kind of computational insights
available to the scientist. Knowledge of brain architecture, far from being
irrelevant to the project, can be the essentialbasisand invaluablecatalyst for
devising likely and powerful algorithms- algorithms that have a reasonable
shot at explaining how in fact the neuronsdo the job .

Levels of Organization

Marr' s three-level division treats computation monolithically, as a single kind

of level of analysis. Implementation and task- description are likewise each
consideredas a single level of analysis. Yet when we measureMarr' s three
levelsof analysisagainstlevels of organizationin the nervoussystem, the fit is
2 .
poor and confusing at best. To begin with , there is organized structure at
different scales:molecules, synapses , neurons, networks, layers, maps, and systems
(figure ). At eachstructurally specifiedstratumwe canraisethe compu-
2.1
tational question: what does that organization of elementsdo? What does it

NeuroscienceOverview
contribute to the wider, computational organization of the brain? In addition,
there are physiological levels: ion movement, channelconfigurations, EPSPs
(excitatory postsynapticpotentials), IPSPs(inhibitory postsynapticpotentials),
action potentials, evoked responsepotentials, and probably other intervening
levels that we have yet to learn about and that involve effects at higher
anatomicallevels suchasnetworks or systems.
The range of structural organization implies, therefore, that there are many
levels of implementationand that eachhasits companiontask description. But

-t. --
if there are as many types of task descriptionsas there are levels of structural
organization, this diversity could be reflected in a multiplicity of algorithms

if
~S
:AND
TEM
/.PAT
C. CENTERS
AND
LOCAL
CIRCUITS

~
In
~
- f\
D. NEURON

.-IlseS
In
"""....
IIIIIIIIII

/\
. \.../ ' -'
r.spons ,
~
~
\

III
A. BEHAVIOR In
.-IlseSIII III III I
out

n8 G or
neurolransmll
Uromo
Jla
c{ Ior "Q j)

~
E. MICROCIRCUITS

"
~dl8-
" "
second. -
~ I ~
:~
(!)Ion _~
s.Wl8l
nadlvily
ger ~ \
~
-1J
-U -. ~.000
-.~Lr -jj
~at~~-
protein

~
F. SYNAPSE

-
-

G. MEMBRANES , IONS
, MOLECULES
....

Figure2.1 Levelsof organizationin the nervoussystem

, ascharacterized
by GordonShepherd
(1988a).

Q, apter2
that characterizehow the tasksare accomplished . This in turn meansthat the
notion of thealgorithmic level is as over-simplified as the notion of the implementation
level.
Note also that the very samelevel of organizationcan be viewed computa-
tionally (in terms of functional role) or implementationally (in terms of the
substratefor the function), dependingon what questionsyou ask. For example,
the details of how an action potential is propagatedmight, from the point of
view of communicationbetween distant areas, be consideredan implementation
, since it is an all-or-none event and only its timing carriesinformation.
However, from a lower structural level- the point of view of ionic distributions
- the
propagating action potential is a computational construct whose
regenerativeand repetitive nature is a consequenceof severaltypes of nonlinear
voltage- dependentionic channelsspatially distributed along an axon.

I.AE -l r...l
:.DIALPREFRor

~-~
OLFACTORY

Figure2.2 A Battened projectionof thecerebralcortexin theright hemisphere of themacaque

monkey. Stipplingindicatescorticalareasimplicatedin visualprocessing . ( Upperleft) Lateral
view of macaque brain. showingvisualareas . (Lowerleft) Medialview of macaque brain. (Reprinted
with pelmissionfromvanEssen andAnderson1990.)

NeuroscienceOverview
 . .. . .................
'; ::::::: .:.::
. .......... ~...~.
.......
: . . . . ... ..
:: .,.
:: :: .
" ......... ..
. ..... ... ....::
...:.. :
: ; ---------............
~: :--...._:..

F
UNiLAMINAR BILAMINAR
ORIGIN ~ T1ON ORIGIN
($ or I) (F, C, orM) (B)

SOF
&
o
c
SUPRA
-
F
GRANULAR

.
LAYER
ASCENDING
(FORWARD
)

LATERAL
1fFRA
-
GRANULAR

DESCENDING
(FEEDBACK
)

Chapter2
Levelsof Processing

The focus for this levels concept is the link between anatomy and what is
representedin the anatomy. As a first pass, it assumesthat the greater the
distancefrom cells responding to sensory input, the higher is the degree of
information processing. Thus the level-rank assignedis a function of synaptic
distancefrom the periphery. On this measure , cellsin the primary visual areaof
the neocortex that respondto oriented bars of light are at a higher level than
cells in the lateral geniculatenucleus(LGN), which in turn are at a higher level
than retinal ganglion cells. Becausethe nature of the representationsand the
transformationson the representationsare still poorly understood, only the
relative level- x is higher or lower than y- rather than the ordinal level-
first, second, etc.- is referredto.
Once the sensory information reaches the cerebral cortex, it fans out
through cortico-cortical projections into a multitude of parallel streams of
processing. In the primate visual system, 2S areasthat are predominantly or
exclusivelyvisual havebeenidentified (van Essenet al. 1991; figure 2.2). Many
(perhapsall) forward projections are matchedby a backward projection, and
there are even massivefeedbackprojections from primary visual cortex to the
LGN. Given thesereciprocalprojections, the processinghierarchy is anything
but a one-way ladder. Even so, by examining the cortical layer into which
fibersproject, it is possibleto find someorder in the information flow. Forward
projections generally terminate in the middle layers of cortex, and feedback
projections usually terminate in the upper and lower layers.3 So far, however,
the function of these feedbackpathways is not established , though the idea
that they have a role in learning, attention, and perceptual recognition is
not unreasonable . If higher areascan affect the flow of information through
lower areas, then strictly sequentialprocessingcannot be taken for granted
(figure 2.3).
The organization typical of earlier sensory areas is only approximately,
4
roughly, and incompletely hierarchical. Beyond the sensoryareas, moreover,
not even that much hierarchy is manifest. The anatomy of frontal cortex and
other areasbeyond the primary sensoryareassuggestsan information organization
more like an Athenian democracythan a Ford assemblyline. Hierarchies
typically have an apex, and following the analogy, one might expect to find a

characteristicpatterns for the cells of origin of different pathways. Bilaminar (B) patterns, shown
on the right , include approximately equal numbersof cells from superAdal and deep layers (no
more than a 700/0- 30% split) and are found to occur with all three types of tennination pattern.
Unilaminar patterns, shown on the left, include predominantly superAdal-layer inputs (S pattern)
which correlate with F-type tenninations, and predominantly in &agranular-layer (I pattern) inputs
which correlate with M -type tenninations. Within this general framework, a number of
variations on a theme can be encountered. Some pathways tenninate primarily in superAdal
layers, but they are grouped with the M pattern becausethey avoid layer 4. Other pathways are
quasi-columnar, but do not include all layers; they are classifiedas a C pattern if the labeling in
layer 4 is neither heavier nor sparserthan in adjoining layers. Filled ovals, cell bodies; angles,
axon terminals. (From Fellemanand van Essen1991.)

NeuroscienceOverview
brain region where all sensoryinformation convergesand from which motor
commandsemerge. It is a striking fact that this is falseof the brain. Although
there are convergent pathways, the convergenceis partial and occursin many
placesmany times over, and motor control appearsto be distributed rather
than vested in a commandcenter (Arbib 1989, Altman and Kien 1989; figure
2.4).
The assumptionthat there is a sensory-processinghierarchy, if only to a first
approximation, affords the possibility of probing the processing stages by
linking variousbehavioralmeasures , suchastask-relative reactiontime (RT) , to
events taking placein the processinghierarchy at different times as measured
by cellular response. To put it more crudely, temporal ordering helps determine
what is causeand what is effect. Accuracy of responseunder varying
conditions can be measured , and both humansand animalsmay be subjects.
This is an important method for triangulating the brain areas involved in
executing a certain task and for determining something about the processing
stagesof the task. For example, on the physiological side, one may measurethe

CNS

Figure 2.4 Model for decision-making in the insect nervous system. In the CNS, stations 1, 2, 3
contain local networks 1, 2, 3. These stations approximate the brain, the subesophageal(SOG),
and segmentalganglia of the locust. The output of each station results from a consensusbetween
the activity of the inputs and the local networks in that station, so the output of each
station is different. The stations are thus linked in several parallel loops, and the output of the
whole system is the consensusof the activity in all the loops. (FromAltman and Kien 1989.)

O1apter2
 delay between the presentationof a moving target and the first responseby
motion-sensitivecells in visual areaMT , and on the behavioral side one may
measurethe responselatency relative to degreesof noise in the stimulus. One
surpriseis that the latenciesfor signalsreachingthe visual areasin the cortex
are so long, relative to the behavioral RT. The latency for MT is about 50-
60 msec, and about 100 msec in inferotemporal cortex. Sincehuman RT to
a complex object is on the order of 150- 200 msec including assembling
the motor response, sending the signal down the spinal cord, and activating
the muscles, this suggests that surprisingly few processing steps intervene
between detection in MT and preparing the responsein the motor cortex,
striatum, cerebellum, and spinal cord. Suchdata help constrain theoriesabout
the natureof the processing.
By way of illustration, considera set of experimentsby William Newsome
and colleagues(1989) in which they show a correlation between the accuracy
of the behavioral responseto motion detection and the spiking frequencyof
single neuronsrespondingto motion stimuli in MT . ( Newsomeet al. 1989) In
the task, tiny dots move randomly on a TV screen. The monkey is trained to
respond as soon as it detectscoherent motion, to either the right or the left.
Across trials, what varies is the number of dots moving coherently and their
direction of motion. The monkey detects direction of motion with as few as
four dots moving coherently, and his accuracyimproves asthe numberof dots
moving together increases . What about the cells in MT? Supposeone records
from a cell that prefersright -going motion. The visual display is set up so that
it is matchedto the cell's receptive Aeld, with the result that the experimenter
has control of the minimum stimulus needed to produce the maximum response
. So long as fewer than four dots move coherently, the cell does not
'
respond. With increasing numbers of dots moving coherently in the cell s
preferred direction, the cell respondsmore vigorously. Indeed, the accuracy
curve displayed in the monkey' s behavior and the spiking-frequency curve
displayedby the single cell are, to a first approximation, congruent(Agure2.5).
This implies, to put it crudely, that the information contained in the cellular
responsesof single sensory neurons and the information contained in the
behavioral responseare roughly on par. It should, however, be kept in mind
that the monkeys were very highly trained on this task and that the sensory
stimuluswas chosento match the optimal responseof eachneuron. In a naive
monkey, there may not be suchclosecorrespondencebetweenthe responseof
the single cell and the overt behavior.
The next phase of the experiment tests whether the information carried
by directionally selective cells found in MT is really used in generating the
response. To do this, Newsome and colleaguespresented left-going visual
stimuli, and at the proper latency-they electrically stimulatedthe column containing
cells preferring right-going visual stimuli. How did the animalbehave?
Would the electrical stimuli demonstrate its effectivenessby overriding, at
least sometimes, the visual stimultl The monkey behaved as though he saw
right -going stimuli; more exactly, the electricalstimulus decreasedthe probability
that the animal would respond to the visual stimulus and increasedthe

NeuroscienceOverview
 correlation . 1 2.8%

IIi)
"~
0
-
~
20

t
zci
(
.
c
0
'
.
.
.
.
0
(
~
N
0
20

;'
~

8
0
.
c
0
:
!
.
0

1.0
u

Spikes per trial

-u
~
.u
0 0.8
.Q c
0
-~
0
Q.,0.6
) 100
10
0.1 1.0Correlation
(%
Figure2.5 (a) Responses of a directionallyselectiveneuron(in visualareaMT) at threedifferent
motion correlationsspanningphysiologicalthreshold . Hatchedbars representresponses to
motionin theneuron's preferreddirection ; solidbarsindicateresponses to motion1800opposite
to the preferreddirection . Sixty trials wereperfonnedin eachdirectionfor eachof the three
correlationlevels.Response distributionsfor a rangeof correlationlevelswereusedto compute
a " neurometric " functionthat characterized the neuron's sensitivityto the motion signal and
couldbe comparedwith the psychometricfunctioncomputedhorn the monkey's behavioral
response. (b) Comparison of simultaneously recordedpsychometric andneurometricfunctions .
Openscircles , psychophysical performance of the monkey; filled circles
, performance of the
neuron.Psychophysical performance at eachcorrelationis givenby the proportionof trialson
whichthe monkeycorrectlyidentifiedthe directionof motion. Neuronalperformance is calculated
Horndistributionsof responses of thedirectionallysensitiveMT neuron . Thephysiological
andpsychophysical datafonn similarcurves , but thedatafor theneuronlie to theleft of thedata
for themonkey,meaningthattheneuronwassomewhatmoresensitivethanthemonkey.(From
Newsome , Britten, and Movmon [1989]. Reprintedby permissionHornNature341: 52- 54.
Copyright@ 1989MaanillanMagazines Ltd.)

OIapter 2
 probability that it would respond as though presentedwith a stimulus in the
' - and hence
opposite direction. This result implies that the cells responses
the information carried in those responses ':'--are behaviorally
significant
(figure 2.6).
During the past hundred years, experimentalpsychologistshave assembled
an impressivebody of RT information, and it is a valuable data base upon
which neuroscientistsmay draw. Thus consideralso a set of studiesby Requin
and colleagues(Requinet al. 1988, Riehleand Requin 1989). In the first stage,
'
they measuredthe monkey s RT where the task was to makea wrist flexion in
a certaindirection and by a certainamountasindicatedby a signal. There were
basicallythree conditions: the monkeyswere precuedor not, and if they were
precued, the cue indicatedeither the direction or the extent of the movement.
Precuingwas found to have a large effect on the RT but only a slight effect on
the movement time, showing that precuing has its major effect on programming
and preparing for the movement, rather than on the speedof execution
of the movement. Additionally , if the advancecuespecifiedwhere but not how
much, the RT was shortenedmore than if the cue specifiedhow much but not
where. This suggeststhat information about extent of movement cannot be
efficiently incorporateduntil the systemknows the direction of the movement.
In the secondstage, Riehleand Requininvestigatedthe electrophysiological
properties of cells in the primary motor cortex ( MI) and the premotor cortex
(PM ). They found execution-relatedneurons, which were more commonin MI ,
and preparation-related, directionally selective neurons, which were more
common in PM. This cohereswith other physiological data, and implies that
PM probably involves an earlier stageof processingthan does MI , sincePM
hasmore to do with preparingfor the movementthan with executingit. Moreover
, within the class of preparation-related cells in PM, they found two
subclass es: those related to programming the musclemovements, and those
related to preprocessingthe general componentsof the movement program.
This is another instance of researchthat narrows down hypotheses about
relative order of processingand the structures involved in a distinct aspect
of processingby establishing behavioral reaction times and by correlating
thosedata with specificresponsesof cells.!

3 STRUCTURE AT VARIOUS LEVELSOF ORGANIZA nON

Identification of functionally significant structure at various spatial scalesin

nervoussystemsproceedsin partnershipwith hypothesesabout a given struc-
'
ture s role in the nervous system's performanceand the mannerin which that
structure's own subcomponentsare organizedto constitute the mechanismsto
carry out that role. To be sure, functional architectureat various spatialscalesis
all part of one integrated, unified biological machine. That is, the function of a
neuron dependson the synapsesthat bring it information, and, in turn, the
neuron processes information by virtue of its interaction with other neuronsin
local networks, which themselvesplaya particularrole by virtue of their place
in the overall geometry of the brain.

NeuroscienceOverview
 RecePtivefield
Pre' LED
.
+
FP
0;
~
ci
,
,"
'
i,
"0,',0 .s
SUO
StImulusaperture .
Null LED . - . stimulation

0
;~
G - - E) no stimulation

FIxatIon poInt ~ L - -

d
\d0 "
,,
0' '
,
VIs. stImulus

Elect . stImulus
~ L - -

.N
d o
'
,""
,
b"'
Q
,
d-0-~ "
~ L -

Target LEDs rL -

3-2.
_,.
0
0-100(% 10
) 20 30
Eye posItIon ~ I . - . - . . stimulation
G - - 0 no stimulation
TIme I I I
TI T2 TJ
. .
B I second D Correlation

Figure2.6 Microstimulationin corticalareaMT blasesperceptualjudgmentsof motion. (A)

Schematic diagramof theexperimental protocolshowingthespatialarrangement of thefixation
point (FP), receptivefield (shaded ), stimulusaperture(thick circle), andresponselight emitting
diodes(LEOs). (B) Schematic drawingillustratingthe temporalsequence of eventsduring a
microstimulation trial. At time T1 the fixationpoint appeared , and the monkeytransferredits
gazeto thefixationpoint, asindicatedby thedeflectionin theeyepositiontrace.At time T2 the
visualstimulusappeared , andthe train of electricalstimulationpulsesbegan.The monkeywas
requiredto maintainfixationfor 1 secuntil time T3. The fixationpoint, the visualstimulus , and
the microstimulation pulseswere turnedoff at time T3' and the target LED turnedon. The
monkeythenindicatedits judgmentof motiondirectionby makinga saccadic eyemovementto
oneof the two responseLEOs. (Right) The effectof microstimulation on performance for two
stimulationsitesin areaMT (C andD). Theproportionof decisionsin thepreferreddirectionis
plottedasa functionof thepercentcorrelationin themovingdotsduringthestimuluspresentation
'
(positivecorrelationvaluesindicatemotionin the neurons preferreddirection). In half the
trials(closedcircles), microstimulation wasappliedsimultaneously with the visualstimulus ; the
other trials(opencircles ) containedno microstimulation . The shift in the curvescausedby the
microstimulation is equivalentto adding7.7%correlateddots(0 and20.1%( 0). (FromSai7man ,
Britten, andNewsome1990. Reprintedby pennission fromNature346: 174- 177. Copyright@
1989MaanillanMagazines Ltd.)

O1apter2
 Accordingly , which structures really constitute a level of organization in the
nervous system is an empirical , not an a priori matter . We cannot tell , in
'
advance of studying the nervous system , how many levels there are, nor what
is the nature of the structural and functional features of any given level . Some
techniques used to study various levels will be surveyed in the appendix . In
this section , seven general categories of structural organization will be discussed
. In fact , however , the count is imprecise, for several reasons. Further
research may lead to the subdivision of some categories , such as systems, into
Aner-grained categories , and some categories may be profoundly misdrawn
and may need to be completely reconfigured . As we come to understand more
about the brain and how it works , new levels of organization may be postulated
. This is especially likely at higher levels where much less is known than at
the lower levels.

Systems

To standardizereferencesto brain locations, prominent landmarks, including

major gyri , fissures, and the major lobes have been labeled (figures 2.7, 2.8).
Using tract-tracing techniques,neuroanatomistshave identified many systems
in the brain. Somecorrespondto sensorymodalities, suchas the visual system;
others, for example, the autonomic system, respectgeneralfunctional characteristics
. Yet others, suchas the limbic system, are difficult to define, and may
turn out not to be one system with an integrated or cohesivefunction. The
componentsof thesesystemsarenot neatly compartmentalizedbut are distributed
widely in the brain and are connectedby long fiber tracts. For example, a
particularbrain systemfor long-term memory may involve suchdiversestructures
as the hippocampus, the thalamus, the &ontal cortex, and basalforebrain
nuclei (Mishkin 1982). In this respectbrain systemscontrastquite vividly , and
perhapsdiscouragingly, with systemsdesignedby an engineer, where components
arediscreteand functionsare compartmentalized .
One of the earliest systemsconceptswas that of a reflex arc, such as the
monosynapticreflex in the knee-jerk responseSherrington 1906; figure 2.9).
The pathwaysof somereflexes havenow beentracedin great detail; examples
are the vestibulo-ocular reflex, which stabilizesimageson the retina when the
head is moving (Robinson 1981), and the gill withdrawal reflex in Aplysia,
which hasbeena focusfor researchinto the molecularmechanismsof plasticity
(Kandelet al. 1987). The reflex arc is not a usefulprototype for brain systemsin
general- or even, it appears, for most reflexes, suchas the stepping reflex in
the cat, or the nociceptive reflex (withdrawal of limb &om a painful stimulus).
Take, for example, the smooth pursuit system for visually tracking moving
targets, where one pathway originates in the retina, leads to the lateral geniculate
nucleus(LGN ), to the cortex and through distinct visual topographic
areas, down to the pons, and eventually to the oculomotor nuclei (Lisbergeret
al. 1987). (Seechapter6.) Despite the machine-like quality of smooth pursuit, it
is to someextent under voluntary control and dependson expectationaswell

NeurosdenceOverview
O1apter2
~ 5aiewJ
(

'
Figure2.8 Major gyri andfissuresof the humancerebralcortex. (Left) View &om above(donal
). ( Right) View &ombelow(ventralor inferior, aspect
aspect ). (CourtesyHannaDamasio .)

as the visual stimulus. Behaviorsmore sophisticatedthan simplereflexes probably

exploit more complexcomputationalprinciples.
In this regard, two important features of brain systems should be mentioned
. First, there are almost always reciprocal (feedback) connectionsbetween
brain areas, at least as rich in number as the feedforward connections.
For example, the recurrentprojectionsfrom the visual cortical areaVI back to
the LGN are about ten times as numerousas those from the LGN to the VI .
Second, although the simplemodelsof reflex arcssuggestthat a single neuron
may be sufficient to activate the neuron on which it synapses , in fact a large
number of neurons are almost always involved, and the effect of any single
neuron on the next is typically quite small. For example, an important feature
in the visual system is that input from a specificneuron in the LGN generally
makesrelatively weak synaptic contactson a large population of cortical cells
rather than a strong synapticeffect on just one or a few neurons( Martin1984).
This implies that cortical neuronsrely on a convergenceof many afferents, and
correlationsbetweenpairs of neuronstends to be relatively weak. 6 There may
be interesting exceptionsto this; for example, chandeijercells in cortex make
inhibitory connectionson the axon hillocks of their targets, and they may, as
single cells, have a strong, decisiveeffect on their target cells. Another exception
is the strong influencethat single climbing fibers have on single Purkinje
cellsin the cerebellum.

Topographic Maps

A major principle of organization within many sensoryand motor systemsis

the topographic map. For example, neuronsin visual areasof cortex, such as

NeuroscienceOverview
 To. . , ~ .;:
. . ~

Figure2.9 Schematic diagramof thepathwaysfor the stretchreflex . Stretchreceptonin mulde

spindlesread to changes in lengthof themuscle , andafferentAbencarrythisinformationalong
thedonal rootsto the spinalcordwherethey synapse on extensormotoneurons , whichextend
the knee, and inhibitory interneurons, which reduceactivity in motor neuronsthat produce
contractionsof the antagonisticAexormuscles . Both of theseactionscombineto producea
coordinatedexpression of theknee . Thisinformationis alsoconveyedto higherbrain
-jerkreflex
centen, whichin turncanmodifythereflexbehaviorthroughdescending pathwaysto thespinal
cord. (FromKandel1985.)

VI , are arranged topographically , in the sensethat adjacent neurons have

adjacentvisual receptive fields and collectively they constitute a map of the
retina. Becauseneighboring processingunits (cell bodies and dendrites) are
concernedwith similar representations , topographic mapping is an important
meanswhereby the brain manages save on wire and also to share wire
to
( Mead1989). It is significantthat the mapsare distorted, in the sensethat some
regions of the body surfaceoccupy larger regions of cortex than others. The
fovea, for example, occupies a r.elatively large part of VI , and the hands
occupy a relatively large area of the somatosensorycortex. In visual area
MT of the macaque , which contains many neurons selectivefor direction of
motion, the lower half of the visual field has greater representationthan the

Otapter 2
Figure 2.10 Schematicdrawing of the multiple representationsof the body surfacein the primary
somatic sensorycortex of the owl monkey. Becausethe cortex of the owl monkey is relatively
flat, most of the body representationis located on the surfacerather than in the convolutions
found in the speaesof most other primates. (A ) Two representationsof the hand are shown in
areas3b and 1. (B) The hand of the owl monkey is innervated by the median and ulnar nerves,
which have different territory on the ventral surface(Bl) and are representedin adjacentareasof
cortex in eachof the two maps(B2). The topographical organization of the cortical map for the
ventral surfaceof the hand is highly ordered (B3) in both areas. Cortex devoted to the ventral
surfaceis indicated in white ; that devoted to the dorsal surface, in dark shading. D . to Ds, digits;
P. to P~, palmar pads; I, insular pad; H, hypothenar pads; T , thenar pads. (From Kandel and
Schwartz 1985).

upper half. This makessensebecauseit is the lower half of the visual field
where hand skills- searchingfor termites, picking up lice, and so forth -
requirethe greatestacuity ( Maunselland van Essen1987.)'
In the visual systemsof monkeys, physiologists have found about 2S distinct
areas,most of which are topographically mapped.s A similar hierarchyof
multiple topographic maps is found for body location in the somatosensory
system ( Kaaset at. 1979; figure 2.10), for frequency in the auditory system
(Merzenich and Brugge 1973), and for muscle groups in the motor system
(Ferrier 1876, Asanuma1973). One possibleexceptionis the olfactory system,
but even odors may be spatially organized at the level of the olfactory bulb
(Stewartet at. 1979). To someextent the different sensorymapscanbe distin-

NeuroscienceOverview
guished by differences in the fine details in the laminations of neurons (see next
section ) and their cellular properties , but often these are so subtle that only
physiological techniques can distinguish bouridaries between different cortical
areas.
Some brainstem structures , such as the superior colliculus , also display this
organization . The cerebellum appears to have patches of partial maps, though
the principles do not seem clear, and these areas may not be maps in any real
senseat all. Some areas seem to lack a strong topographic organization , and for
other areas the topographic organization is quite complex , for example the
basal ganglia (Selemon and Goldman - Rakic 1988 ). Cortical areas anterior to
the central sulcus seem sparser in topographic maps, but research may show
that what they map are abstract, not sensory , representations , and hence such
maps cannot be discovered by methods used to establish response patterns to
peripheral stimuli . In bat auditory cortex there are topographic mappings of
abstract properties such as frequency differences and time delays between
emitted and received sounds, properties that may help the bat to echolocate
prey (Suga et al. 1984 ), and in the bam owl internal spatial maps are synthesized
from binaural auditory inputs ( Konishi 1986 , Knudsen et al. 1987 ). There
are some areas of cortex , such as association areas, parietal cortex, and some
parts of frontal cortex , for which it has not yet been possible to find properties
that form orderly mappings . Nonetheless , projections between these areas
remain topographic . For example , Goldman - Rakic ( 1987 ) has shown that in the
monkey projections from parietal cortex to target areas in the prefrontal cortex
, such as the principal sulcus, preserve the topographic order of the source
neurons.
Maps of the sUlface of the body in the brain are formed during development
by projections that become ordered , in part , through competitive interactions
between adjacent fibers in the target maps (see chapter 5). Some of the neurons
undergo cell death during this period , and with the possible exception of
olfactory receptors , no new neurons are formed in the mature mammal (Cowan
et al. 1984 ). However , competitive interactions between neurons continue , to
some extent , even in adulthood , since the territory in cortex devoted to a
-
particular part of the body sUlface can shift as much as 1 2 cm, but not much
farther , weeks after injury to sensory nerves or after excessive sensory stimulation
(Pons et al. 1991 ). Thus , regions in somatosensory cortex that are silenced
following denervation of a sensory nerve will eventually become responsive
to nearby regions of the body . It is not yet known how much of this rearrangement
is due to plasticity in cerebral cortex , or perhaps in subcortical structures
that project to cortical maps. Auditory maps, particularly in the superior col -
liculus , are also modifiable both in development , and in the adult following
partial deafness (King and Moore , 1991 ). Nonetheless , this evidence , and further
evidence for synaptic plasticity summarized below , make it difficult to
" "
think of the machinery in the adult brain as hardwired , or static. Rather, the
brain has a remarkable ability to adapt to changes in the environment , at many
different structural levels and over a wide range of time scales.

O\ apter2
 Imm

Figure 2.11 Cross - section through monkey striate cortex using cresyl violet to stain cell bodies .
Laminations are clearly visible ; the layers are numbered at the left . W , white matter . Deeper
layers of the burled fold of cortex are shown in the lower part of the figure . (From Hubel and
Wiesel 1977 .)

Layers and Columns

Many brain areas display not only topographic organization , but also laminar
organization (Agures 2.11, 2.12). Laminae are layers (sheets) of neurons in
register with other layers, and a given lamina conforms to a highly regular
pattern of where it projects to and from where it receives projections . For
example , the superior colliculus receives visual input in superficial layers, and
in deeper layers it receives tactile and auditory input . Neurons in an intermediate
layer of the superior colliculus represent information about eye movements .
In the cerebral cortex , specific sensory input from the thalamus typically projects
to layer 4, the middle layer , while output to subcortical motor structures
issues from layer 5, and intracortical projections originate chiefly in (superficial )
layers 2 and 3. Layer 6 mainly projects back to the thalamus (see Agure 2.3 ).
The basal ganglia do not have a laminar organization , but rather a patchwork
of islands which can be distinguished by developmental and chemical markers
(Graybiel and Hickey 1982 ).
As well as the horizontal organization seen in laminae, cortical structures
also display vertical organization . This organization consists in a high degree
of commonality between cells in vertical columns , crossing laminae, and is

Neuroscience Overview
 ( A )

AFFERENTS E FIELD TYPES EFFERENTS

RE C Emv

. .
, .

'
.
,
U & :
.

. .
. .

m
.
.
,

Iva b
,

I Vc

. . '
.
, ,

VI

Simple

n lr = = -
U U u

0dIer Lateral
Superior

cortical co Uicu Ius

geniculate

areas nudeus
4
~

A I Inin8eA C I Inin8eA

Lateral
geniculate

nucleus

( 8 )

U & :

Iva b
,

I Vc

VI

Geniculate Cortex Cortex

Lateral
Superior

axons
co I Iku Jus

geniculate

nucleus

Figure2.12 Schematic diagramof corticalconnectionsin thecat. (A ) Distributionof inputsfrom

, showingthat geniculateaxonsprojectto differentcorticallami-
layersof the lateralgeniculate
naeasa functionof layerof originin thegeniculate. Corticalneuronswith similarreceptivefield
propertiesclustertogetherin particularlamina. The origin of fibersleavinga regionof cortex
variesas a functionof target. (8) Schematic arborizationpatternsof the main cell types in
laminaeI- VI. (After GilbertandWiesel1981.)

dtapter 2
reflected both anatomically in terms of local connections between neurons
(Martin 1984, Lund 1987) and physiologically in terms of similar response
properties (Hubel and Wiesel 1962). For example, a vertical penetration of an
electrodein visual cortex revealscellswhich sharea preferencefor stimuli with
the sameorientation (e.g ., a bar of light oriented at about 200 from the horizontal
). Another vertical penetration nearby will show cells which prefer a
different orientation. Inputs and outputs are also organizedin columns, suchas
the oculardominancecolumnsin VI , and inputs into the principal sulcuswhich
alternatebetweenparietalprojectionsfrom the sameside and projectionsfrom
the principal sulcusin the opposite hemisphere(Goldman-Rakic1987).
Typically , the vertically organized connectivity patterns do not result in
columns with sharp boundaries, and the responseproperties tend to vary
" "
continuously acrossthe cortex. Hence the expression vertical column may
be slightly misleading. Thus for cellsin visual areaVI , orientation varies over
the cortex smoothly, save for some fractures and singularities (Blasdel and
Salama1986), and a similar organizationcanbe found in areaV2 (Swindaleet
al. 1987), which receivesa topographically mappedprojection from VI . There
are, however, placeswhere vertical, cross-laminar columns with quite sharp
boundariesare seen, for examplethe ocular dominancecolumns in layer 4 of
area VI and the "barrels" in the rodent somatosensorycortex, where each
barrelcontainscellspreferentiallysensitiveto stimulationof a particularwhisker
( Woolseyand van der Loos 1970) (figure 2.13). Sharpanatomicalboundaries
are, however, the exception rather than the rule. Also, the spatial scale of
columnar organization can vary from about 0.3 mm for ocular dominance
columnsto 2S.umfor orientation columnsin monkey visual cortex.
Topographic mapping, columnarorganization, and laminaeare specialcases
of a more general principle: the exploitation of geometric properties in information
processingdesign. Spatialproximity may be an efficient way for biological
systemsto assemblein one place information neededto solve aproblem
. To considera simple case, supposeit is necessaryto comparedifferences
between stimuli at neighboring locations, where comparisonrequiressignals
be brought together. Then topographic organization may achieve this efficiently
while minimizing the total length of the connections. This is desirable
sincemost of the volume of the brain is filled with axonal processes, and there
are limitations on how big the brain canbe as well as temporal tolerancesthat
must be met. Lateralinhibitory interactionswithin the spatialmapsareusedto
make comparisons , enhancecontrast at borders, and perform automatic gain
control. Mutual inhibition within a population of neurons can be used to
identify the neuron with the maximum activity , a type of winner-take-all circuit
(Feldmanand Ballard 1982). (SeealsochapterS, last section.)

Local Networks

Within a cubic millimeter of cortical tissue, there are approximately 10Sneurons

and about 109 synapses , with the vast majority of thesesynapsesarising
from cellslocatedwithin cortex (Douglasand Martin 1991) (figure 2.14). These

Neuroscience Overview
Figure 2.13 (A ) Snout of a mouse; the vibrissae (whiskers) are marked by dots. ( 8) Sections
acrossthe somatosensorycortex that receiveinput &om the mout . Eachof the rings or "barrels"
correspondsto an individual vibrissa, and are spatially organized to preservethe neighborhood
relations of the vibrissae(0 . ( Reprintedwith permission&om Woolsey and van der Loos 1970.)

OIapter 2
Thalamus

Figure2.14 Schematic diagramof a microcircuitin the cerebralcortexthat may be repeated

and
again again . Threepopulationsof neuronsinteractwith eachother: inhibitory(GABA) ceDs,
shownwith solid synapses ; and excitatoryceDs(open synapses ) representing(i) superficial
(P2 + 3) and(ii) deep(PS+ 6) layerpyramidalcells. Eachpopulationreceivesexcitatoryinput
from the thalamus , which is weaker(dashedline) to deeppyramidalceDs. ( Reprintedwith
permission Douglaset at. 1989.)
from

local networks have been very difficult to study owing to the complexity of
the tangled massof axons, synapses , and dendritescalled the neuropil. Nevertheless
, somegeneralfeaturesof local networks are beginning to emerge. For
example, the orientation tuning of cells in VI must emerge&om nonoriented
inputs and activity in local networks in ways that we are just beginning to
understand(Fersterand Koch 1987).
Most of the data available on local networks are basedon single-unit recordings
, and to achievea deeperunderstandingof the principles governing
networks, it will be necessaryto monitor a large population of neurons (see
Appendix for recording techniques). Even a local network involves many cells,
but only small populations canbe studiedby exhaustivesequentialrecordings
&om single cells. Consequently, we run the risk of generalizing&om an atypical
sample, and of missing circuit properties that can be inferred only &om a
richer profile. Therefore, to understandthe principles of local networks, much
more work must be done to detennine the dynamical traffic within a larger
population of cellsover an extendedperiod of time (figure 2.15).
Computer simulations may help to interpret single-unit data by showing
how a population of cells could representproperties of objects and perfonn
coordinate transformations. For example, network models of spatialrepresentations
have been constructedthat help to explain the responseproperties of
single cells in parietal cortex (Andersen and Mount castle 1983, Zipser and
Andersen1988; figure 2.16). Another network model hasbeenusedto explain
how the responsesof single neurons in visual cortex areaV 4 could compute
color constancy,(Zeki 1983, Hurlbert and Poggio 1988). Network simulations
canalso suggestalternativeinterpretationsfor known responseproperties. For

NeuroscienceOverview
 uoie
uoQ

- . . -
neuron showing one of its inputs to a dendrite, one to the cen body , and one of its axonal
contacts. (Reprinted with permissionfrom Dudal TheNeurobiologyof Memory: Concepts ,
, Findings
and Trends[ 1989]. Copyright ~ Oxford University Press.)

example, there are certain oriented cells in VI whose responsesumrnateswith

the length of the slit or edge of light up to the borders of the receptive ReId,
but then the responsediminishes as the length increases . This property, called
" end- " has related to the extraction of the ID curvature
stopping , recently been
of contours (Dobbins et ale 1987) and the 2-D curvature of shapesin
shadedimages(Lehky and Sejnowski 1988). An example of this approachis
given in chapter4 on visual processing.

Neurons

'
Ever since Cajal s work in the late nineteenth century, the neuron has been
taken as an ele~ entary unit of processingin the nervous system (figure 2.17).
" "
In contrast to Golgi, who believed neuronsformed a continuous reticulum,
or feltwork, Calal arguedthat neuronswere distinct, individual cells, separated
from each other by a spatial gap, and that mechanismsadditional to those
operating intracellularly would have to be found to explain how the signal

Chapter2
 -
.

!
D

- - - - - - - - - - - - - - - - - - - - - - - - - - - - -
~ - - - - - - - - - - - - - - - - - -

".,".,0
:,.,'~
, '',",,,"'"'.",".'",:..,..''",""
.",., ,
D

..
'.
.'.
..
.. -.
"
.. .
...
!
;-
" ""
. "'
I

.",-.,,...,,,., .-."..... . "

.0.0
- ,:,."., '"
.., ,
,,.:,..!I'I1I'.,-":.,.,".';;

,,
,,,
u " - ,
'110 .-'11
.~l~
,.to_ :-~i. ~~~-~~"r,.-r
,IIII.'.'..,'..'..'..'.'.'.'.'.'.''tOO .'-,'.' ---_8 tO O
-
Figure
parietal 2cortex
.1must of 6an ration of
the
awake
, -umonkey
single
behaving nit
technique
.The used
to
animal the
study
fixated
as
~ ofneuron
response
targetin
lightthe
at
a
place
8TMa . . -

series
shown of
herepositions
.At each on the
fixationwith
screen
, its
a
position head
squarewas .
fixed
Theresults
Bashedobtained
for
1sec
at at
100two
above
the are
positio
of
point
fixation
single . Recordings
trial ,and each from
a
small
nick neuron
single
made on are
the shown
line below
the
saeen
the
represents.Each
line
of
an
discharge by a
repres
the
impuls
neuron
figure . The
shows impulses
the res were
~ summed
nses
for in the
fixation
to the on
histograms
leftthe
and bottom
and
down
, ,The
panel
the
leftside
of
right
side the
show
the
responses
neurons in for fixation
parietal to
cortexthe
has and
right up. This
fields
receptive and
that
areothera
to show
that
experimen
retinal
specific this
butclass
the
location
, of
degr
of
activation
position of of
theeye the neuron
.(Salso to
eeavisual
stimulus
4,section
chapter within
10 the field
.)(FAndersen
rom andis
modula
receptive
Mount ,)the
by
castle
1983
NeuroscienceOverview
Figure2.17 Examples
of neurons thevarietyof shapes
illustrating in differentareasof thebrain.
fromKuffler
( Withpermission , NicholisandMartin[1984 ]. FromNeuron to Brain
. Sunderland
MA: SinauerAssociates
.)

'
passedHorn neuron to neuron. Physiological studies have borne out Cajal s
judgment, though in some areassuch as the retina, syncytia of cells that are
electricallycoupledhave beenfound (Dowling 1987). As it turns out, theseare
rather more like the structuresGolgi predictedbecausethe cells are physically
" "
joined by conducting gap junctions. Theseelectricalsynapsesare fasterand
more reliablethan chemicaltransmission,but aremore limited in flexibility .
There are many different types of neurons, and different parts of the nervous
systemhave evolved neuronswith specializedproperties. There are Ave
generaltypes of neuronsin the retina, for example, eachwith a highly distinctive
morphology, connectivity pattern, physiological properties, and embryo-
logical origin. In recentyears, moreover, physiologicaland chemicaldifferences

O1apter2
 Active Inactive

I ,

1 &

~
.

.
-
. - -
~ ~
: :
t , . . : : : :
-

Figure2.18 Inhibitoryand excitatorysynapses on a neuron. (A ) The inhibitory postsynaptic

potential(IPSP ) meansthat the postsynapticcell hyperpolarizes (droppingfrom - 70 mV to
- 72 mV), and the excitatory
postsynapticpotential(EPSP ) meansthat the postsynapticcell
depolarizes (from - 70 mV to - 67 mY). (8) The EPSPwastriggeredabout1. 3, andS msec
aftertheonsetof theIPSP . (C) Thesubsynaptic conductance changes occurringwhenexcitatory
andinhibitory synapses areactivatedsimultaneously (left) and whenonly the excitatorysynapse
is activated(right). (FromSchmidt[1978). Fundamentals -
. Berlin: Springer
of Neurophysiology
Verlag.)

have been found within classes . For example, 23 different types of ganglion
cells(whose axons project to the brain through the optic nerve) and 22 different
types of amacrinecells (which provide lateral interactions and temporal
differentiation) have been identified (Sterling et al. 1983). There are seven
general types of neuronsin the cerebellumand about 12 general types in the
neocortex, with many subtypes distinguishableby their chemicalproperties
suchasthe neurotransmittersthey contain. The definition of a neuronaltype is
somewhat arbitrary, since judgments are often made on the basis of subtle
morphological differences, which can be graded rather than categorical. As
more chemicalmarkersare found, however, it is becomingclearthat the diversity
of neurons within cerebral cortex has been vastly underestimated . On
anatomical and immunocytochemicalcriteria, therefore, the number of subtypes
of cortical neuronsis probably between50 and 500 (Sereno1988).

NeuroscienceOverview
 Dendrites

Myelinatedaxon
Axonhillock
Presy ~

terminal

. .

"

. .

~
- . . . .

~ ~ .A
L ~~ ~ ___--~ L__~v.....J l..._ ...- -..-v - -- -----...-
Graded
EPSP Trigger
-
: Conducted all - or - none spike
all or-none (conduction of spike to next cell )
sDike initiated

Figure 2.19 Summary diagram showing the location on a motor neuron of various electrical
events. In many neurons, dendrites and cell bodies respond with graded EPSPsor I PSPs; the
action potential is triggered in the axon hillock and travels undiminisheddown the axon. (From
Thompson 1967.)

On the basisof their effects, neuronsdivide in two generalclasses : excitatory

and inhibitory (figures 2.18, 2.19). The effect of an excitatory signal is to
increasethe probability that the postsynaptic cell fires, and the effect of an
inhibitory signal is to decreasethat probability (figure 2.19). Some neurons
also have modulatory effectson other neurons, principally by releasingpep-
tides or monoamines(see section 4). Another useful classificationconcerns
projections: somecellsramify only within a confinedareasuchasa column , for
example stellate cells in cortex; other neurons, such as pyramidal cells, have
long-range projections out of an area, where the route goes via the white
matter ratherthan directly through the cortex itself. Researchon the properties
of neuronsshows that they are much more complex processingdevicesthan
previously imagined (table 2.1). For example, dendrites of neuronsare themselves
highly specialized , and some parts can probably act as independent
processing units (Shepherdet al. 1985, Koch and Poggio 1987).

Synapses

Chemicalsynapsesare found in nervous systemsthroughout phylogeny, and

they are a basicunit of structurethat hasbeenhighly conservedduring evolution
. A synaptic bouton has a surfacearea of a few squaremicrometersand
forms a highly stereotypedappositionwith the postsynapticmembrane,which
itself is highly specialized(figure 2.20). Synapsesare the primary gatewaysby
which neuronscommunicatewith one another, and they consistof specialized

O1apter2
Table 2.1 Selected biophysical mechanisms , possible neural operations they could
implement , and computations they might help perform
Biophysical Mechanism
Actionpotentia ]I initiation
Repetitive spiking
activity
Action potential
conduction
Conduction failure at
axonal branchpoints
Chemicallymediated
synaptic transduction
' mediated
Electrically
synaptic transduction
Distributed excitatory
synapsesin dendritic tree
Interaction between
excitatory and (silent)
inhibitory conductance
inputs
Excitatory synapseon
dendritic spinewith
calciumchannels
Excitatory and inhibitory
synapseson dendritric
spine
Quasi-active membranes
Transmitter regulation of
voltage- dependent
channels(M -current
inhibition )
Calciumsensitivity of
cAMP - dependent
phosphorylation of
potassiumchannelprotein
Long~ stanceaction of
neurotransmitter
FromKochandPoggio(1987
).
NeuroscienceOverview
Neural Operation
Analog OR/ AND one-bit
analog-to- digital converter
Current-to- &equency
transducer
Impulsetransmission
Temporal/ spatialfiltering
of impulses
Nonreciprocal two- port
" "
negative" resistance
"
Sigmoid threshold
Reciprocalone-port
resistance
Linearaddition
AnalogAND-NOT, veto
operation
Postsynaptic modification
in functionalconnectivity
LocalAND-NOT
"
inhibition"
presynaptic
Electricalresonantfilter
analog
Differentiationdelay
Gaincontrol
Functionalconnectivity
Modulating and routing
transmissionof
infonnation
Example of Computation
Long-distance
communicationin axons
Opener musclein
cray6sh
Coupling of rod
photoreceptorsto
enhancedetection of
signals
~, P catretinalgangliol1
cells
Bipolar cells
Directional-selective
retinal ganglion cells
Disparity -selective
cortical cells
Short- and long-term
information storage
Enabling/ disabling
retinal input to
geniculate X-cells
Hair cells in lower
vertebrates
Midbrain sites
controlling gain of
retinogeniculate
transmission
Adaptation and
associativestorageof
information in Aplysia
 -
-
~
&
D. ~

extracellular
matrix

Dendritic
sPecialization

Figure2.20 Schematic diagramof a synapseon a dendriticspine. Denseprojectionsin the

presynaptic membrane are surroundedby vesiclesthat presumablycontainneurotransmitter
molecules. This morphologycharacterizes a type I synapse, which is excitatory. Type II synapses
(not shown) have Battenedvesiclesas viewed in the electronmicroscopefollowing
glutaraldehyde fixation, andtheyareofteninhibitory. (FromGershonet al. 1985.)

presynaptic structures for the release of neurochemicalsand postsynaptic

structuresfor receiving and responding to those neurochemicals . Evidenceis
accumulating that signaling between neurons at synapsescan be selectively
altered by experience(Alkon 1984). Other, structuralcomponentsof neurons
might also be modified through experience,suchas the shapeand topology of
dendritesas well as the spatial distribution of membranechannels(Purvesand
Voyvodic 1987).
Our understandingof the nervous systemat the subcellularlevel is changing
rapidly, and it is apparentthat neuronsare dynamic and complex entities
whose computational.properties cannot be approximatedby memoryless response
functions, a common idealization. It remainsan open scientificquestion
how the integrity of memoriesthat spandecadescanremainintact if the neural
substrateis as fluid as preliminary reports indicate, especiallyif , as it seems ,
networks of neuronsboth processand store information.

Q, apter2
 -b
"~
e.Li-ay
,:1=
"
of Excitable

Molecules
Membranes
-p Aqueous
pore

Figure2.21 Working hypothesisfor a voltage-gatedchannel

. SunderlandMA: SinauerAssociates
.)
. The transmembrane
shownwith a pore that allowssodiumions to flow betweenthe extracellular
sidesof themembrane whenthegateis open. ( Withpermission
proteinis
andintracellular
fromHille [1984]. Ion;c Channels

The integrity of neurons and synapses depends on the properties of membranes

and the internal cytoskeleton of the neuron . The membrane serves as a
-
barrier a few nanometers ( 10 9) thick separating the intracellular and extracellular
aqueous compartments . The membrane itself is a two -dimensional fluid
medium in which integral membrane proteins and other molecules form associations
. Some integral membrane proteins have an important role inmaintaining
the ionic milieu inside and outside the cell. For example , membrane
9
proteins that serve as ion channels can be voltage sensitive , chemically activated
, or both . They may thus permit or prevent the passage of ions across the
membrane, which in turn can affect the propagation of a signal down the
length of the axon or neurotransmitter release at the presynaptic terminal
(figure 2.21 ). In a sense, the membrane allows the intracellular compartment of
a neuron to respond selectively to extracellular signals, and it is this selec-
tivity that endows different neurons with specialized information -processing
capabilities . Axon membrane typically contains channels and conductances

NeuroscienceOverview
that permit it to spikewhen depolarizationreachesa certain threshold. Exactly
how dendrite membraneworks is much lesswell understood. Dendrite spiking
has been seen in the cerebellum (Uinas ' and Sugimori 1980), and the
conventional wisdom according to which axon membraneis " active" while
dendrite membraneis " passive" is undoubtedly a simplification that obscures
the subtle, complex, and computationally critical respectsin which dendrite
membraneis active.
Electricalsignaling in neuronsis achievedby ionic currentswhich are regulated
by ion channelsand ion pumps in the cell membrane. Signaling between
neuronsis mediatedby neurotransmitterreceptorsin the postsynaptic
membranethat respondto particularneurotransmittermoleculesby transiently
and selectively changing the ionic conductanceof the membrane. There are
also receptor moleculesalong the membraneoutside of the synaptic site that
appear to be functional, but their role is not known (Somogyi et al. 1989).
In addition, somereceptorscan activate one or more second-messengermolecules
that can mediate longer-term changes(figure 2.22). Second-messengers
in neurons can be activated by more than one receptor. Hence there is a
network of interacting chemicalsystemswithin a neuron, which can itself be
considereda chemicalparalleldistributed processor.

4 A SHORTLIST OF BRAIN FACTS

A central part of the basicstrategy for figuring out how a novel device works
is reverse engineering. That is, when a new cameraor chip appearson the
market, competitors will take it apart to find out how it works. Typically, of
course, they alreadyknow quite a lot about devicesof that generalkind, so the
problem can be manageable . Although we have to use reverseengineeringto
study the brain, our starting point is much further back, inasmuchas we know
so little about devicesof that generalkind. From our vantagepoint, the brain is
essentiallya bit of alien technology, and henceit is especiallydifficult to know,
among the facts availableto us, which are theoretically important and which
are theoretically uninteresting. We may actually misunderstandsome aspects
of brain organization and as a consequencebe blocked from having some
important insight into mechanismscrucial for cognition. For example, some
distinctionsmadein grossanatomy may turn out to concealcloserelationships
betweendistant .brainregions, or it may turn out that the functional properties
of somesynapsesin the centralnervoussystemare very different from peripheral
synapsesin autonomic ganglia and neuromuscularjunctions, which have
beenvery well studied.
Sincethis chapter looks at neuroscienceagainsta backgroundof computa-
tional aims, it seemsappropriateto raisethis question: what are the most basic
structuralfeaturesrelevant to neuralcomputation? It goes without saying that
many more constraintswill be relevantin the context of a specificproblem, but
we presentthese13 asa kind of prolegomenonto problemsgenerally. Short of
having formally conducteda proper survey, we conjecturethat the following
baker's dozen are among those likely to find their way on to a must-know list,

Otapter2
 ND
<
(N
+
c
@-
+
IIn
2
C
\~
~
P
F @
)
J
'--~+n
/L.
B
C
I
T
1
:
"
\
~
R
P
.
H
@
B
-C
+
-
Sa PK
/ J, , -
@Other
Effects
BIosynU18sis

Transport

Insertion

Storage

Transport

Figure2.22 Summaryof someof the mainbiochemical mechanisms that havebeenidentified

at chemicalsynapses . A - E, Long-tenn stepsin synthesis , transport , and storageof neurotransmitters
andneuromodu1ators ; insertionof membrane channelproteinsandreceptors , and
neuromodu Jatoryeffects. 1- 12, thesesummarize the morerapidstepsinvolvedin immediate
signalingat thesynapse. 1P3, inositoltriphosphate; CaMn, Ca2+/cal moduli n- dependent
protein
kinaseII; DAG, diacylglycerol ; PK, proteinkinase ; R. receptor; G, G protein; AC, adenylate
. ( Reprinted
cyclase with pennission &om Shepherd 1988.)

NeurosdenceOverview
DISTANCE
I I- . II I I I I II I I I I I
-
10.10 10 9 - -- 6
10 10- 3 100 103
. Amp .m IJm mm m
ANGSTROM MICROMETE:RMilliMETER METER

I-6I I 10 I I UI103
(MICRON)

TIME I 10 I-3I I 100 IUI UI

IAsec msec sec I HOURI
\ DAY
I MINUTE
MilliSECOND
SECOND
Figure2.23 Logarithmicscalesfor spatialand temporalmagnitudes . Bracketsindicatethe
. ( Reprintedwith pennissionfrom Shepherd
scalesespeciallyrelevantto synapticprocessing
1979.)

- - - - - - - - -

~
( : J

~ /
I / I ~
" ,

ObjectDiscrimination Landmark Discrimination

Figure2.24 Two behavioraltasksthat distinguishbetweenthe functionsof the inferiortemporal

(IT) andposteriorparietal(PP) cortex. (Left) IT lesions , causea severeimpainnentin
, in black
learningto discriminatebetweentwo objectsbasedon their features , but the lesionsdo not
affectspatialcapadties . (Right) PPlesionscauseanimpainnentto spatialtasks , suchasjudging
whichof two identicalplaquesis closerto a visuallandmark(cylinder), but do not affectobject
discrimination learning. (FromMishkin , UngerleiderandMacko[1983J . Objectvisionandspatial
vision: two corticalpathways 6: 414- 417.)
. Tmrdsin Neurosciences

although we understandvery well that opinion can diverge in considerable

and surprising ways, and also that a current list will undoubtedly be quickly
outdated (for comparablelists, see Crick and Asanuma 1986 and Shepherd
1988). (Seefigure 2.23 for scalesof magnitudes.)
1. Specializationof FunctionThere is specializationof function in different
regions of nervous systems. This is a ubiquitous and critical featureof nervous
system organization, seen.in animalsfrom the lowly leechto the human. The
specializationenjoyed by regions more distant from the periphery, such as
orbitalfrontal cortex of humans, is difficult to determine, though by using a
convergenceof techniques, including lesions, staining, single-cell recording,
evoked potential, and developmentaldata, the range of likely possibilitiescan
be narrowed (figure 2.24). Specializationso characterizedis actually a large-
grain feature of an area, basedon the statistical distribution of cell response
propertiesand the major input and output pathways. Thus VI , for example, is
referred to as a visual area and S1 as a somatosensoryarea. At a finer grain,

Otapter2
 however , the specialization of areas is consistent with the existence of atypical
cell types and connectivity . Thus while the preponderance of tested cells in VI
are indeed visually tuned , there exist some ceMscoding for nonvisual signals,
such as eye movement .
2. Numbers: Neurons and SynapsesThe estimated number of neurons in the
human nervous system is about 1012; the number of synapses is about 1015.
The rat brain has about 1010 neurons , and about 1013 synapses. In 1 mm3 of
cortical tissue there are about 105 neurons and 109 synapses. A handy rule of
thumb is 1 synapse/ .um3. A single neuron may have thousands or tens of
thousands of synapses. Stevens ( 1989 ) has calculated that the number of synapses
per neuron for a piece of cortex 1 mm thick from a cat or a monkey is
4.12 x 103. The main exception to this is the primary visual cortex of primates
, where cells are more densely packed and the number of synapses is
about 1.17 x 103 for a piece of cortex 1 mm thick .
3. Numbers: Connectivity ( Who Talks to Whom) Not everything is connected
to everything else. Each cortical neuron is connected to a roughly constant
number of other neurons , irrespective of brain size, namely about 3% of
the neurons underlying the surrounding square millimeter of cortex (Stevens
1989 ). Hence, although the absolute number of input lines to a .cortical neuron
may be quite large , cortical neurons are actually rather sparsely connected
relative to the population of neurons in a cell ' s neighborhood . Most connections
are between , not within , cell classes (Sereno 1988 ). Forward projections
to one area are generally matched by recurrent projections back to the area of
origin .
4. Analog Inputs/ Discrete Outputs The input to a neuron is analog (continuous
values between 0 and 1), and a neuron ' s output is discrete (either it spikes
or it does not ), though some neurons may have analog outputs . Whether a
neuron has an output is governed by a threshold rule ; that is, whether the cell
spikes depends on whether the integration of the inputs exceeds a certain
threshold . The profusion of input lines to a single neuron probably represents
sensible computational and engineering design for a network of neurons with
these properties (Abu -Mostafa 1989a). 10
S. Timing : General Considerations Getting the timing right is an essential
feature of nervous systems and, more particularly , of analog computation
(Mead 1989 ). Whether and how dendritic signals traveling soma-wards will
interact depends on the time of their arrival at the common node. The magnitude
of signals eventually reaching the axon hillock depends on such interactions
. In perception , the time scale of the computation must be matched to the
time scale of events in the external world , and in motor control it must be
matched to the time it takes for the body parts to move ( Mead 1989). When
outputs of different computational components need to be integrated , the
time scales of the various processors contributing values must also match. In
short , the system has to operate in real time . Hence nervous systems must be
architecturally rigged so that when a process takes time , it takes the right
amount of time .

NeuroscienceOverview
 ( A)

From sympathetic trunk

above 7th ganglion
From 7th and 8th

spinal nerves

mV
m V

:
:

: :~ :: : : :: : ::: : : t L:: ~ f ~ : : : :: = : :: : :: : .J
0 t I..; ~ : :: : : ~ 0
0 20 40 6Omsec 0 2 4 6 8eec

V
mV ( 0) Slow ( E) Late slow epsp
epsp ~
~ ~
" " ~ "
- "
' - ~
- '
- " " - - ~
: ~ -
! j
V
0 0
~ 1~ ~ ~ ~ sec J ~ ~ ~ ~ ~ min

Figure2.25 Fourtypesof synapses from the sympathetic ganglionof the frog. (A ) Innervation
of a sympatheticneuronin the ninth ganglionof the paravertebral chainof the bullfrog; the
diagram shows the separationof the cholinergic(ACt ) and noncholinergic(LHRH ) innervation .
(B) A singlepreganglionicstimulusproducesa fastEPSP . (C) Repetitivestimulationproducesa
slow IPSPlastingabout2 sec; the fast EPSPis blockedwith a nicotinicblockingagent. ( 0)
Repetitivestimulationalsoproducesa slowEPSP whichoccursafterthe first two responses and
lastsabout30 sec. (E) The late slow EPSP , producedby stimulatingpreganglionic fibers, lasts
morethan5 minafterrepetitivestimulation . ( Withpermission fromKufRer , NicholisandMartin
[1984]. FromNeuronto Brain.Sunder landMA: Sinauer Associates .)

6. Timing: ParticularValuesAn action potential (spike) lasts about 1 msec.

Synaptic transmission, including electrotonic conduction in dendrites, takes
about 5 msec. Synapticpotentials can last from a millisecondto many minutes
( Kuffler1980) (figure 2.25). Transmissionvelocity in myelinatedaxonsis about
meters/ sec; in unmyelinatedaxons it is less than 1 meter/ sec. These
aregeneralranges, not precisevalues.
7. Cell-ta-cell EffectsThe effect of an individual synaptic input on a postsynaptic
cell is weak, amounting to 1%- 5% of the firing threshold. There may
be some important exceptionsto this trend, such as. the strong effects of an
individual synapseof a chandeliercell or a basket cell in the cerebralcortex
(Martin 1984).
8. Firing PatternsDifferent types of neurons have different firing patterns
(figure 2.26). Someneuronsin the thalamushave multiple intrinsic firing patterns
, and the particularpattern displayedon a given occasionis a function of

Otapter2
 A -spiking
Regular

-- mV
-J. . 150
-spiking
B Fast .!::::.J3nA
50ms

50 mV
~ 3 nA
~
25 ms

Repetitive Bursting

. J - L

50 ms
Figure2.26 Differencesin intrinsicfiring patternsof corticalneurons
. (A ) Whenstimulatedwith
a suprathreshold stepof depolarizingcurrent, reguJar-spikingneuronsrespondwith an initial
high-frequencyspikeoutput that rapidly declinesto muchlower sustained&equendes . Intracellular
voltagesaredisplayedin the top trace, injectedcurrentstepsin the bottom trace. (8)
Undersimilarconditions , fast-spikingcellsgeneratehigh frequencies that aresustained
for the
durationof the stimulus
. (0 Repetitiveintrinsicburstingto a prolongedstimulus . Meaninterburst
frequencywasabout9 Hz. (FromConnonandGutnick[1990]. Intrinsicfiring patternsof
diverseneocorticalneurons . Trendsin Neuroscimas 13: 98- 99.)

the cell' s recentdepolarizationor hyperpolarizationhistory (Uinas and Jahnsen

1982). The ionic conductancesof some cells, for example in the brain stem,
endow those cells with oscillatory properties. Sucha cell may act as a pacemaker
or asa resonator(respondingpreferentially to certainfiring hequencies )
(Uinas 1988). Most neuronsare spontaneouslyactive, spiking at random intervals
in the absenceof input. Different neuron types have different characteristic
spontaneousrates, ranging from a few spikesper secondto about so spikes
per second.
9. Receptive : Sizeand Center
Fields -SurroundOrganizationUnder the classical
definition, the receptivefield is that region of the sensoryfield from which an
adequatesensorystimuluswill elicit a response.In the somatosensorysystem,

NeuroscienceOverview
 ON-CENTER
CELL

stimulus ~ L -

I IIIIIIIIIII
center
I
~I IIIII
surround
~
center
and
II II IIIII
surround

~
stimulus
center
-CCELL
OFF
II II II II
11111
ENT
L-

II 111111
surround
I 1111111111111
center
andI I III I
surround
II III I
Figure2.27 Two typesof circularcenter- surroundreceptivefieldsin theretina. Whenthelight
is shoneon the centerof the receptivefield, the on-centerceOrespondsvigorously, the off-
centerceOis silent. When the light is shonein the annularsurround , the oppositeeffectis
achieved . Underdiffuseilluminationof both the centerand the surround , both ceOsrespond
.
weakly ( Withpermission &om Coren Ward [1989 ]. and
Sensation Perception, Jrd ed. Copyright
~ 1989 HarcourtBraceJovanovich , Inc.)

the receptivefield sizevariesover the body surface: thosefor the fingertips are
smallerthan those for the palm of the hand, and very much smallerthan those
for the arm. Receptivefields of cells in higher areasof visual cortex tend to be
much larger than those in the earlier stages(one sixth of a degreein the foveal
region of VI , comparedto valuesranging from 10 to the whole visual field in
inferotemporal cortex). Retinal ganglion cells (cells carrying signals from the
retina) have what is called a center -surroundorganization(figures 2.27, 2.28).
This organization comesin two variations: (1) a stimulus in the center of the
'
cell s receptivefield excitesit , but a stimulusin an areasurroundingthe receptive
" "
field inhibits it . This arrangementis known as on-center/ off-surround.
(2) The oppositearrangement , namely, a centralstimulusinhibits but a surround
" "
stimulusexcitesthe cell, is known as off-center/ on-surround. Off-centercells
respondmaximally to dark spots, while on-centercells respondmaximally to
light spots. The information carriedby the ganglion cells pertainsto the comparison
between the amount of light falling on the center of the field and the
averageamount of light falling on the surround, not absolutevaluesof light
intensity at the transducer. A center-surround organization is also evident in

O1apter2
 !
~ ~
c
!
I
C
0.
,
U
-
Q
s~~)~):
pun

JO
UO

Aq e
.

pun

UO
Jo

.0~
i~
U<"
>
~
.~
JOlda

~
l~
~-f-~ - "
~
I
l-
\ --"
\."
,'-"
\~J 3
~J~ I:II
I":~ ,J'\.--'-,--<
III .It/>
,\.,"---."ii
\.I'Z
.c0
~
"
~ij~~=(~-'.~
.
uo

JO

UO
Jllod
.I
ar81

NeuroscienceOverview
 .\.I./.\
DOTS
CENTER
IN
MOVE
OPTIMUM
DIRECTION ..-..r.:I-,-_"._
-J',..-..

NOIJ
BACKGROUND

\""./
DIR
CTION

~~I:,I,
TYPEI
HCMT32B z
DOTS
CENTER 000 000000 0
-
to
MOVE 00
00: - :0000
0 ...... 0 .c

I
I
BACKGROUND
DOTS 0 0 ~~ : 0 ~
..
00000
STATIONARY
0000000000 U
~
. .
'
"./ .\.
...c
Q0~.
3

. .
/ -
I,
-~,
eo-,
1-,-
20,~
e-,o,
C
! , ,
6d, ,
120
- ,
I.(J'-'8
,-1 ,-20 , ,
,-1DIRECTION
-" , \
, ~
-
, .
, DIRECTIO
OF OFMOVEM
DOTS
6 00
C1
OF 60
"
MOVEMENT
OOTS
CENTER
OF 120
-
Figure 2.29 Responseof a neuron with antagonistic direction-selective surround. (Left) The cell
responds vigorously when the dots in the center of the stimulus move (shown above) in the
preferred direction but the dots in the surround are stationary. Negative percentagesin the
graph indicate inhibition relative to the level of spontaneousactivity . ( Right) The same cell
respondsvery differently to its preferred direction of motion in the center when the dots in the
'
surround also move in the cell s preferreddirection. (FromAllman et al. 1985.)

the receptive fields of somatosensoryneurons in the thalamus and cortex

( Mountcastle1957) (figures2.27, 2.28).
10. Receptive Fields Eventsoutside the classicalreceptivefield of
: Nonclassical
a cell havebeenfound to modulateselectivelythe responsesof the cell ( Nelson
and Frost 1978, Allman et al. 1985) (figure 2.29). The effectsare selectivesince
they vary as a function of the type of surround stimuli. Nelson and Frost
(1985) reported an inhibition as well as a highly spedfic form of facilitation of
the responsesof orientation-tuned cellsin visual cortex of catsasa nonclassical
field effect. Some area 17 cells that were normally responsiveto a vertical
bar in their receptive fields showed enhancedresponseswhen distant11area
17 cells, co-oriented and co-axial to the first, were experimentally stimulated.
Zeki (1983) has shown that certain wavelength- dependentneuronsin V4 are
influencedby the color balancein the surround. The surroundeffectsof cellsin
the middle temporal( MT) area, where receptivefieldsaretypically S - 10 , can
extend 40 - 80 (Allman et al. 1985). Receptive fields are almost certainly
more dynamical than previously assumed . For example, repeatedstimulation

Otapter2
Figure2.30 Alterationof corticalmapsafterhabitualstimulation. (A) The experimental
protocol
showingthefingertipsstimulatedby therotatingdisk. (B) Map in thesomatosensorycortex
of the left handof the monkeybeforethe stimulationexperiment. Stippledareacorrespondsto
fingertips2, 3, and 4. (C) Map of the sameregion after the stimulationexperiment. (From
Merzenichet al. 1990.)

to the Angertipsresults in an expansionof the regions of thesomatQsensory

cortex whose neuronshave receptive fields in the Angertips(figure 2.30). Recent
experimentsin in VI of visual cortex also suggeststhat receptive fields
'
are labile in that a cell s receptivefield may expandwhen its preferredareaon
the retina is lesioned(Gibert and Wiesel, in press).
11. Specificand NonspecificSystemsIn addition to the specific system projecting
to the neocortexvia the thalamus, suchasis seenin the visual, auditory,
and somatosensorysystems, there are five sourcesof widely projecting neurons
eachassociatedwith a specificneurotransmitter, which may play important
roles in the sleep- dreaming- waking cycle, in memory, and in awareness
and attention. The five are as follows: the locus coeruleusin the brain stem
(norepinephrine), the raphe nucleusin the midbrain (serotonin), the substantia
nigra in the midbrain (dopamine), the nucleusbasalisin the basal forebrain
(acetylcholine), and specialgroups of cells in the mammillary region of the
hypothalamus(GABA ) (figure 2.31).

NeuroscienceOverview
Chapter2
 ~
a AUTOCRINE b PARACRINE d NEUROTRANSMISSION

-~0
B
S
e NEUROENDOCRINE

~
:-.
zw
.
.
.
.
. . .
.
. .
~ .
: . ; : . . .

~ ~

C ENDOCRINE

NE
.A
..A
...
~ TARGET
CELLS
Figure 2.32 Methods of communication of the honnonal and nervous systems. Although auto-
cine honnones (a) act on the ceOthat releasesthem and paracine honnones (b) act on adjacent
cells, most honnones are in the endocrine system and act on ceUs or organs anywhere in the
body . Endocrineglands (c) releasehonnone moleculesinto the bloodstream , where they come
in contact with receptors on target ceUs, which recognize the honnones meant to act on those
cells and pull them out of the bloodstream. Neurons (d) communicate by releasing neurotransmitters
close to target ceUs. In neuroendocrineaction (e) a neuron releasessubstancesthat
act ashonnones directly into the blood. (Reprintedwith permissionfrom Snyder 1985.)

12. Action-at-a-distanceSomeneurotransmittersmay be releasednot only at

a synaptic site, but may also be dumpedinto the extracellularspaceto have an
action at a nonsynapticsite somedistancefrom the point of releaseOanet al.
1978) (figure 2.32). Originating in the endocrine system, hormones, such as
estradiol, canalso reachneuronsafter traveling through the circulatory system
and canalter neuralactivity .
13. ParallelArchitectureThe brain appearsto be highly parallel in that there
are many parallel streamsof input for a given function. For example, in the
monkey two parallel streamsfrom the retina, starting with different types of
ganglion cells, project to two distinct sets of layers of the lateral geniculate
nucleus- the parvocellularand magnocellularlayers, respectively- which in
turn project to distinct sublaminaein layer 4 of cortical areaVI of the visual
cortex (Hubel and Livingston~ 1987, Livingstoneand HubeI1987 ). The streams
are not cleanly segregated, however, and there are probably interactions at
every stage(Schilleret al. 1990, Logothetis et al. 1990).

Selected Readings
Abeles
, M. (1991). Corliconics
: NeuralCircuitsof the CerebralCoria. Cambridge
: Cambridge
UniversityPress
.

Neuroscience Overview
01angeux, J.-P. (1985). Neuronal
Man. Oxford: Oxford UniversityPress .
Oturchland, P. S. (1986). Neurophiiosophy
: TOWRrdII Unified of theMind-Bmin.Cambridge
. Sdence ,
MA: MIT Press .

Dowling, J. E. (1987). TheRth'nR: An ApprOR

ChRbltPRrlof theBmin. Cambridge
, MA: Harvard
UniversityPress .
Groves,P. M., andG. V. Rebec(1988). lntrodudion
to Biologi
C41l , 3rd ed. Dubuque
Psychology , IA:
WInoC. Brown.
Hall, Z. W. (1991). MoltculRrNeurobiology
. Sunder
landMA: Sinauer
.
Hubel, D. H. (1988). Eye
, VisionRndBrllin. New York: Freeman
.
rod, M. (1985). TheBminMRchine
Jeanne . Cambridge
, MA: HarvardUniversityPress
.
KandeLE., J. Schwartz
, andT. M. Jessel
Leds. (1991). Principles
of NtUrRlScience
, 3rd ed. New
York: Elsevier
.
Kelner,K. , andDE . Koshland
, eds. (1989). Molecules
to Models: AdORnasin Neuroscience -
. WRSh
, DC: AmericanAssociationfor theAdvancement
ington of Science.
, S. W., J. G. Nicolis, andA R. Martin (1984). FromNeuronto Bmin:A CellulRrApproRChto
KufHer
theFunctionof theNm1OUS , 2nded. Sunder
S.vstem land, MA : Sinauer
.
le Vay, S., andS. B. Nelson(1991). Columnarorganizationof the visualcortex. In TheNeuml
&sis of VisURlFunction, ed. J. R. Cronly-Dilion. London
: Maanillan.
, I. B., and L K. Kaczmarek
Levitan (1991). TheNeuron
: CellRndMoltculRrBiology
. Oxford:
Oxford UniversityPress .
, G. M. (1987). Neurobiology
Shepherd .
, 2nded. Oxford: Oxford UniversityPress
, G. M. (1990). 5,vnRpticOrgRniZRtionof theBmin, 3rd ed. Oxford: Oxford University
Shepherd
Press
.
White, E. L (1989). CortiC41l
Circuits
. Boston .
: Birkhauser

Seleded Journals and Reviews

CurrentOpinionin Neurobiology in neuroscience
. (CurrentBiology) Reviewpaperson sub6elds .
; Quarterlyjournal( MITPress
of CognitiwNeuroscience
Journal ). Articleson systemsneuroscience
with emphasis .
on cognitiveprocessing
. Quarterly journal (Saunden). Each issue is on a spedal topic in
SeminRrs in Neuroscience
neuroscience.

Trendsin Neurosaen &tS. Monthly journal (Elsevier). Contains brief but very useful reviews of
specialtopics and is a good sourceof up- to- date l~ cicl . cesto the literature.
. ( WorldScientific). Contains discussionsof conceptualissues.
in Neuroscience
Concepts
. Palo Alto , CA: (Annual Reviews). Comprehensivereviews of the
Annual Reviewof Neuroscience
literature.

Otapter2
This excerpt from

The Computational Brain.

Patricia Churchland and Terrence J. Sejnowski.
© 1994 The MIT Press.

is provided in screen-viewable form for personal use only by members

of MIT CogNet.

Unauthorized use or dissemination of this information is expressly

forbidden.

If you have any questions about this material, please contact

cognetadmin@cognet.mit.edu.