R E V I E W S
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |7 9 3
Neocortical neurons are not randomly distributed in
the cortical sheet, but are arranged in layers (layers
I–VI) that connect to different cortical and subcortical
regions1–3.In rodents,a neocortical column ofabout
0.3 mm in diameter contains roughly 7,500 neurons4,5
(100 neurons in layer I;2,150 in layer II/III;1,500 in
layer IV;1,250 in layer V and 2,500 in layer VI).Most
neocortical neurons (70–80%) are excitatory pyramidal
neurons1,3,6,7,which have relatively stereotyped anat-
omical, physiological and molecular properties1,8.
The remaining 20–30% of neocortical neurons are
interneurons,mostly inhibitory interneurons,which
have diverse morphological,physiological,molecular
and synaptic characteristics3,6,8–16.
Despite this diversity,inhibitory interneurons have
many common features, some of which distinguish
them from pyramidal neurons. First, most mature
inhibitory interneurons have aspiny dendrites3,6,17.
Second,interneurons can receive both excitatory and
inhibitory synapses onto their somata3,17,18.Third,the
axons ofinhibitory neurons usually arborize within a
cortical column and can project laterally across columns,
but do not typically project down into the white matter
to contact distant brain regions19.Indeed,neocortical
interneurons in general are also called ‘local circuit
neurons’ to reflect the restriction oftheir axonal and
dendritic arbours to the neocortex20.Fourth,different
types ofinhibitory neuron seem to be especially capable
oftargeting different subdomains ofneurons (dendritic
regions,soma or axon)14,21.
Interneurons can also be excitatory.The spiny stellate
cell (SSC) is an important type ofexcitatory (glutamater-
gic) interneuron that has a star-like dendritic arboriza-
tion with a high density ofspines around its soma.These
cells are found only in layer IV of primary sensory
areas22,23;receive excitatory inputs from specific thalamic
nuclei3;and relay this information to layer II/III24–26,indi-
cating that they are specialized to process thalamic input.
SSCs share many characteristics with pyramidal neurons,
but lack a prominent apical dendrite27;instead,they have
only a partial apical-like dendrite that can reach layer III.
There are also excitatory peptidergic interneurons,
including a subgroup ofbipolar interneurons — small,
oval cells with axons and dendrites that stretch vertically
in a narrow band across all layers.
Inhibitory interneurons,which use GABA (γ-amino-
butyric acid) as their transmitter,vary greatly intheir
somatic,dendritic and axonal morphologies (FIG.1).
Dendritic morphology is the most variable feature and
cannot reliably define the type ofinterneuron.However,
the axonal aborization can reveal the anatomical identity
ofan interneuron because interneurons seem to be part-
icularly specialized to target different domains ofneu-
rons,different layers ofa column and different columns.
INTERNEURONS OF THE
NEOCORTICAL INHIBITORY SYSTEM
Henry Markram*,Maria Toledo-Rodriguez*,Yun Wang‡,Anirudh Gupta*,Gilad
Silberberg* and Caizhi Wu§
Abstract | Mammals adapt to a rapidly changing world because of the sophisticated cognitive
functions that are supported by the neocortex. The neocortex, which forms almost 80% of the
human brain, seems to have arisen from repeated duplication of a stereotypical microcircuit
template with subtle specializations for different brain regions and species. The quest to unravel
the blueprint of this template started more than a century ago and has revealed an immensely
intricate design. The largest obstacle is the daunting variety of inhibitory interneurons that are
found in the circuit. This review focuses on the organizing principles that govern the diversity of
inhibitory interneurons and their circuits.
Fédérale de Lausanne,
1 Bungtown Road,
Cold Spring Harbor,New
Correspondence to H.M.
(PV).A calcium-binding protein
that can act as an endogenous
buffer in certain neurons.
(CB).A calcium-binding protein
that might function as a calcium
(CR).A calcium-binding protein
that can be used as a marker of
7 9 4 |OCTOBER 2004 |VOLUME 5
R E V I E W S
that results from convergent innervation by many
basket cells.On the basis ofdifferences in their axonal
and dendritic morphologies,basket cells can be divided
into three main subclasses:large basket cells (LBCs),
small basket cells (SBCs) and nest basket cells (NBCs).
Basket cells typically express many neuropeptides and
the two calcium-binding proteins, PARVALBUMIN (PV)
and CALBINDIN(CB) (FIG.3).
Large basket cells.LBCs are the classic basket cells.They
have large,aspiny,multipolar dendrites and expansive
axonal arborizations that can inhibit neurons in upper
and lower layers and in neighbouring and distant
columns2,3,17,28,38,39,41.LBCs are therefore the primary
source oflateral inhibition across columns within the
layer that contains their somata. The local axonal
arborization ofLBCs is sparse,has a low bouton density
and,uniquely,tends to branch sharply (FIG.1;online
supplementary information S1(table)),often giving it a
stick-like appearance.The somato–dendritic morphology
is often multipolar,but can be bitufted,pyramidal or
bipolar. LBCs can express CB, PV, neuropeptide Y
(NPY), cholecystokinin (CCK) and occasionally
somatostatin (SOM) and CALRETININ (CR).They never
express vasoactive intestinal peptide (VIP) (FIG.3).
Small basket cells. SBCs are aspiny, soma-targeting
interneurons with local, dense and highly varicose
axonal arborizations that seldom send axons beyond
a cortical layer or column2,28,38,39,42,43(see online supple-
mentary information S1 (table)). Their somato–
dendritic morphology can be multipolar,bitufted or
bipolar,with different tendencies depending on the
layer;SBCs in layer IV are often multipolar and in layer
II/III are more often bitufted or bipolar.SBCs can also
be readily distinguished from LBCs by their frequently
branching and ‘curvy’axons (FIG.1;online supplementary
information S1(table)).Occasionally,a few collaterals
extend out of the local axonal cluster.SBCs form the
highest number ofsynapses on pyramidal neurons (FIG.4;
online supplementary information S2(table)).They
differ from other basket cells in that they express VIP
(FIG.3).A special subtype ofSBC,the clutch cell,is found
in layer IV ofthe visual cortices ofcats and monkeys43.
These are medium sized,multipolar cells that typically
produce curvy axonal collaterals with large bulbous
terminals that appear to ‘clutch’ the somata of their
Nest basket cells.NBCs were frequently reported in the
past and sometimes referred to as interneurons with
‘irregular arborizations’30,44–46,but were only recently
shown to be a distinct class ofsoma-targeting cell15,38.
The name ‘nest’ arises because oftheir birds’-nest-like
appearance.NBCs seem to be a hybrid of LBCs and
SBCs,but have a local axonal cluster more like SBCs
and less frequent branching and longer axonal colla-
terals with a lower density ofboutons,more like LBCs
(FIG.1;online supplementary information S1(table)).
NBCs do not typically express CR and never express
This is a crucial issue as many studies,especially recent
studies that lack a thorough anatomical background,
have confused cell types by relying only on dendritic
features.In other words,the multipolar,bipolar and
bitufted classifications ofinterneurons cannot be used
alone to identify a cell type (see section on morpho-
logical properties). However, on the basis of their
axon targeting, interneurons can be functionally
divided into axon-targeting, soma- and proximal
and tuft-targeting interneurons14,21, and also into
columnar, intralaminar–intercolumnar and inter-
laminar–intercolumnar interneurons. The relative
percentages ofeach interneuron type vary in different
species, brain regions and layers3,12,28–37. Here, we
review the characteristic features and connections of
the most common types of inhibitory interneuron,
with a particular emphasis on quantification obtained
in the rat somatosensory cortex.
Morphological properties of interneurons
Basket cells.About 50% ofall inhibitory interneurons
are basket cells (FIG.2).Basket cells specialize in targeting
the somata and proximal dendrites ofpyramidal neu-
rons and interneurons2,3,31,38–40(FIG.1),which places them
in a unique position to adjust the gain ofthe integrated
synaptic response.The term ‘basket cell’comes from the
basket-like appearance around pyramidal cell somata
Cajal Retzius cell
Dendrite-targeting cellsDendrite- and tuft-targeting
Soma- and proximal dendrite-targeting cells
Spiny stellate cellNest basket cell
Small basket cell
Figure 1 |Anatomical diversity of neocortical neurons. Scheme summarizing the main
anatomical properties of neocortical inhibitory interneurons. Each neuron type has a different
coloured soma; dendrites, red; axons, blue lines; axonal boutons, blue dots. Spines are omitted
for clarity. Neurons are orientated with the pia facing upwards and white matter downwards.
Some interneurons have a prominent, vertical dendrite directed towards the white matter.
Inhibitory interneurons are mainly distinguished by the structure of their axonal arbour (see text)
and typically innervate selective domains ((peri-) somatic, dendritic or axonal) of their target cells.
Modified, with permission, from REF.167© (2002) MIT Press.
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |7 9 5
R E V I E W S
horizontally in layer I for millimetres3,8,12,19,56,57to inhibit
tuft dendrites in neighbouring and distant columns,
providing the only source for cross-columnar inhibition
via layer I from layers II–VI.MCs target not only the
most distal dendrites,but also proximal dendrites,peri-
somatic dendrites and somata56.Infragranular MCs can
also selectively target layer IV56. MCs are therefore
unusual in that they target multiple domains and multi-
ple layers.They most often have bitufted morphology
with a more elaborate dendritic tree than most
interneurons.The axons ofMCs are also unusual in that
they form spiny boutons.MCs always express SOM and
never express PV or VIP (FIG.3).
Bipolar cells.Bipolar cells (BPCs) are small cells with
spindle or ovoid somata and narrow bipolar (most
often) or bitufted dendrites that extend vertically
towards layer I and down to layer VI3,58–60(FIG.1).Their
axon commonly emerges from one ofthe primary den-
drites and forms a narrow (<50 µm) band that crosses
all layers (see online supplementary information S1
(table)).Bipolar neurons can be excitatory by releasing
only VIP, or inhibitory by releasing mainly GABA
(inhibitory BPCs also express VIP).Their bouton den-
sity is low compared with other interneurons,and they
therefore contact only a few cells,mainly on the basal
dendrites ofpyramidal neurons.BPCs occur in layers
II–VI,and typically express CR and VIP (FIG.3).
Double bouquet cells. Double bouquet cells (DBCs)
usually have a bitufted dendritic morphology.Their
special feature is a tight fascicular axonal cylinder3,61–63
that resembles a ‘horse tail’ (FIG.1).The highly varicose
collaterals that form these columnar bundles are unusu-
ally thicker than the axonal main stem and can extend
across all layers.In primates,DBCs seem to be inter-
leaved with pyramidal cells to inhibit their basal den-
drites63.The axons ofDBCs branch frequently to form
higher-order branches and are densely studded with
boutons.DBCs mainly innervate dendrites (spines and
shafts) and are therefore dendritic-targeting cells (see
online supplementary information S1(table)).DBCs
occur in layers II–V,although they seem to be preferen-
tially located in the supragranular layers.They express
CB,have the unique tendency to express CR and CB
together and can also express VIP or CCK,but not PV,
SOM or NPY (FIG.3).
Bitufted cells.Bitufted cells (BTCs) are similar to BPCs
and DBCs in that they usually have ovoid somata and
give rise to primary dendrites from opposite poles to
form a bitufted morphology (FIG.1).However,unlike
the narrow vertical axonal projection ofBPCs,and the
‘horse-tail’ axonal cluster of DBCs, BTC axons have
wider horizontal axonal spans,even across a cortical
column (see online supplementary information S1
(table)).The vertical projection is also less extensive
and crosses mostly to neighboring layers. BTCs are
dendritic-targeting cells14that are found in layers II–VI.
They can express CB,CR,NPY,VIP,SOM and CCK,but
not PV (FIG.3).
Chandelier cells. Chandelier cells (ChCs) are axon-
targeting interneurons14,21,47,48. This targeting could
place ChCs in a powerful position to override all the
complex dendritic integration and somatic gain settings
by ‘editing’ the action potential output49–51.They can be
multipolar or bitufted.Their local axonal clusters are
formed by high-frequency branching at shallow angles,
often ramifying around,above or below their somata
with a high bouton density.The characteristic terminal
portions ofthe axon form short vertical rows ofboutons,
resembling a chandelier47,52(FIG.1).The chandelier-like
appearance seems to become progressively more refined
in ‘higher’ species,with the clearest form (and perhaps
greater abundance) in primates12,53.ChCs have been
found in layers II–VI.They typically express one or both
ofthe calcium-binding proteins PV54and CB (FIG.3).
Similar axon-targeting neurons have been described in
Martinotti cells.Martinotti cells (MCs) are found in
layers II–VI.They specialize in projecting their axons
towards layer I,where they inhibit the tuft dendrites
of pyramidal neurons (FIG.1;online supplementary
information S1(table)).Their axons can also project
Large basket cell
Nest basket cell
Small basket cell
Double bouquet cell
Figure 2 |Different types of interneuron in the layers of somatosensory cortex of juvenile
rats. Note that the percentage of neurogliaform cells might be artificially small because very small
somata were often overlooked during recording.
7 9 6 |OCTOBER 2004 |VOLUME 5
R E V I E W S
Electrical properties of interneurons
Neocortical neurons have various active and passive
properties,so they respond differently when excited
with a depolarizing step current pulse69–72.The class-
ification of these responses has been refined over the
past decade.Originally, all inhibitory interneurons
were described as fast spiking (FS)69,71,but subsequent
recordings revealed other discharge patterns11,73,such
as those of low-threshold-spiking (LTS) cells, also
known as burst-spiking non-pyramidal (BSNP)
cells11,73,74.These show typical burst-like discharges
after a hyperpolarizing pre-pulse and are found in
layer V (which also contains many bursting pyramidal
cells).Some ofthese interneurons have been anatomi-
cally identified as MCs and DBCs. Regular-spiking
non-pyramidal (RSNP) cells discharge in a manner
that resembles regular-spiking pyramidal cells11,75,76.
These cells have been recorded in layers II/III and V,
and some have been identified as MCs, DBCs and
BPCs.Late-spiking (LS) cells,which discharge with a
considerable delay after a depolarizing step,have also
been reported.These cells were found in layers II/III
and V, and some were identified as NGCs11,75.
Irregular-spiking (IS) cells fire an initial burst ofaction
potentials followed by irregularly spaced action poten-
tials,and form a small fraction of interneurons with
bipolar morphology in layers II/III and V9,77.IS cells
have been further divided,according to the duration of
the initial burst,into IS1 and IS2 cells77.The character-
istics of these bursts differ from those of intrinsically
bursting pyramidal cells (TABLE 1).An attempt to map
synapse types between pairs ofdifferent types ofneu-
ron led to the conclusion that the resolution of these
classification schemes was not sufficient to uniquely
identify inhibitory interneurons. A classification
scheme was therefore recently developed that is based
on both the onset and the steady-state response to a
step current injection into the soma15.
Steady-state response types.When divided according to
their steady-state response,interneurons fall into five
groups (TABLE 1; FIG.5):non-accommodating (NAC);
accommodating (AC); stuttering (STUT); irregular
spiking (IS);and bursting (BST).NACs fire repetitively
without frequency adaptation in response to a wide
range ofsustained somatic current injections.The inter-
spike intervals ofconsecutive action potentials during
the steady state either do not change or change mini-
mally,and the discharge frequency increases steeply as a
function ofthe injected current amplitude.Their single
action potentials are very briefand characteristically
have a deep fast afterhyperpolarization (fAHP).ACs fire
repetitively with frequency adaptation and therefore do
not reach such high firing rates as NACs, but some
could be classified as fast spiking (TABLE 1).STUT cells
fire high-frequency clusters ofaction potentials inter-
mingled with unpredictable periods ofsilence (‘Morse-
code’-like discharges) for a wide range of sustained
somatic current injections.The action potentials in a
cluster show hardly any accommodation,and the silent
periods between clusters vary unpredictably in duration.
Neurogliaform cells.Neurogliaform cells (NGCs) are
small,‘button-type’ cells with many fine, radiating
dendrites that are short, aspiny, finely beaded and
rarely branched3,39,64.They form a highly symmetrical
and spherical dendritic field.The axon can arise from
any part of the soma or from the base of a dendrite,
and shortly after its origin,it breaks up into a dense,
intertwined arborization of ultra-thin axons with as
many as ten orders ofbranching (FIG.1).Fine boutons
are distributed on the axonal collaterals to form GABA
synapses onto the dendrites of target cells39. The
molecular characteristics ofNGCs are not well studied.
Layer I interneurons.Virtually all layer I neurons are
inhibitory65,66,and they fall into two categories,which
might reflect different origins65–68.The first comprises large
neurons with horizontal processes,known as Cajal Retzius
cells,which seem to be present mostly during develop-
ment,and are unique to layer I.These multipolar cells can
have various soma shapes, which probably arise from
adaptations to different locations in layer I.Their axons,
which are confined to layer I,can be extensive and typi-
cally have a horizontal trajectory from which extend many
short ascending and some descending terminal fibrils
that are believed to target the terminal tufts ofpyramidal
cells (FIG.1).The second category is a heterogeneous group
ofsmall,multipolar interneurons with varying axonal
arborizations (poor and rich axonal plexus cells).
Large basket cell
Nest basket cell
Small basket cell
Double bouquet cell
Figure 3 | Expression of calcium-binding proteins (CBPs) and neuropeptides in
interneurons. Expression profiles of the CBPs calbindin (CB), parvalbumin (PV) and calretinin
(CR) and the neuropeptides neuropeptide Y (NPY), vasoactive intestinal peptide (VIP),
somatostatin (SOM) and cholecystokinin (CCK) by different morphological and
electrophysiological classes of interneuron. AC, accommodating; b, burst subtype; c, classic
subtype; d, delay subtype; IS, irregular spiking; NAC, non-accomodating; STUT, stuttering.
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |7 9 7
R E V I E W S
cells in supragranular layers; t-BST cells produce a
powerful burst only once at the onset ofthe depolariza-
tion followed by complete cessation ofspiking after the
initial burst due to a powerful sAHP;and i-BST cells
initially burst and then switch to slow accommodation
after the initial burst response,like some pyramidal
neurons (FIG.5; TABLE 1).
it was assumed that a given firing pattern reflected a
certain anatomical type of interneuron. However,
detailed examination of many of the neocortical
interneuron types has shown that a morphologically
identified interneuron can have many discharge behav-
iours8,11,15,38,73,78. TABLE 1shows the mapping between
the different classification schemes,and FIG.6shows the
mapping between the electrical and anatomical types.
A principal component analysis using many specific
electrical parameters can generate clusters that map
differently onto pyramidal cells,MCs and basket cells
(M.T.-R.,B.Blumenfeld and H.M.,unpublished obser-
vations),indicating that higher-order statistics ofthe
electrical properties ofinterneurons might,however,
correspond to specific morphologies.
Classes or a continuum?Electrical classifications pro-
vide a useful means to refer to different types of
response.The proofthat these responses represent dis-
tinct classes and that each class maps onto anatomically
and molecularly distinct types of interneuron is still
lacking.Extrapolating these types to the in vivositua-
tion or assigning them greater significance than a
‘marker’ might also be misleading,because the different
responses are characteristic for highly standardized
experimental conditions — that is,brain slices perfused
with artificial cerebrospinal fluid.In vivoneurons also
receive greater synaptic bombardment and are subject
to neuromodulatory control,which could profoundly
alter their discharge properties79.Nevertheless,under
controlled conditions the response types are useful
markers for describing the microcircuit and under-
standing the relationships between electrical behaviour
and the morphological,molecular and synaptic prop-
erties ofthe microcircuit,regardless ofwhether these
are distinct classes or specific ranges ofresponses.
Electrophysiological diversity results from the combined
activity ofdifferent combinations ofion channels on a
neuron’s membrane (active properties)80and from the
morphology ofthe neuron (passive properties)81.Electro-
physiology, pharmacology, immunohistochemistry,
in situhybridization and gene-expression studies in neo-
cortical regions or single cells have shown that many ion
channels are involved in generating electrical behaviour
in neocortical neurons.Each type ofneuron expresses a
specific combination ofion channels,produces certain
amounts of each channel, uniquely modifies each
channel and distributes them in a characteristic pattern
across the membrane surface to generate a specific type
IS cells discharge single action potentials randomly
throughout the ‘steady-state’phase ofsustained current
injections and have been found only occasionally in
layers II–V77.IS cells tend to show marked accommoda-
tion.BST cells characteristically fire a cluster of3 to 5
action potentials riding on a slow depolarizing wave,
followed by a strong slow afterhyperpolarization
(sAHP).This burst is similar to the classic burst found
in pyramidal cells and not like the transient bursts
found in other interneurons (see below).ACs and NACs
are the most common response types encountered in
the juvenile somatosensory cortex.Stuttering behaviour
is less common, but has been observed in all layers
(II–VI).BST cells have been observed in layers II–V,but
are most commonly found in layer V.
Onset response types.Neurons can be divided into three
subclasses according to the type of onset that their
response to a step depolarization shows.b-NAC cells
initially discharge a cluster of three or more action
potentials,d-NAC cells show a delay before discharge
onset and c-NAC cells have neither bursts nor delays at
the onset (the ‘onset’phase is indistinguishable from the
‘steady-state’ phase).These responses are referred to as
classicalresponses (TABLE 1;FIG.5).All three ofthese ini-
tial response subclasses are also seen in AC and STUT
cells,and the b- and c- subclasses have also been found
for IS cells.b-IS cells have been further subdivided into
those with briefbursts and those with prolonged bursts
(IS1 and IS2 subtypes)77.The BST subclasses are named
differently:r-BST cells repetitively burst in a manner
similar to the ‘chattering’ response ofsome pyramidal
Synapses onto pyramidal cells
Number of contacts
Number of contacts
Synapses from pyramidal cells
Figure 4 |Contact numbers of interneurons onto and
from pyramidal cells. Top, the number of synapses onto
pyramidal cells made by the various interneuron types; bottom,
the number of synapses each type of interneuron receives from
pyramidal cells. Error bars correspond to the standard deviation.
7 9 8 |OCTOBER 2004 |VOLUME 5
R E V I E W S
Revealing candidate genes. The correlation map
revealed many candidate genes that underlie different
electrical properties.For example,in addition to Kv3.1,
Kv3.2and PV,the expression ofanother delayed recti-
fier,Kv1.6,and ofthe high-threshold Ca2+channel gene
Caα1G and its auxiliary subunit,Caβ4,is also highly
correlated with fast spiking,whereas the expression
ofCR,Caα1I,Kv2.2,HCN4and SK2is negatively cor-
related with fast spiking.The precise pacing ofspiking
to minimize accommodation in fast-spiking neurons
is also mostly correlated with the expression of the
hyperpolarization-activated channels HCN1 and
HCN2and ofKvβ1— an auxiliary subunit ofthe Kv1
gene family that transforms these delayed rectifiers into
transient A-type channels with important pacemaker
Three gene clusters.A cluster analysis ofco-expression
also revealed three main classes ofion-channel expres-
sion, which surprisingly mapped around the three
calcium-binding proteins (PV,CB and CR) that are
expressed in neocortical neurons8,10,11(FIG.3).The ion-
channel genes that are co-expressed with CR (the ‘CR
cluster’) include SK2,Kv3.4,CR and Caα1B;the ‘CB
cluster’ includes CB,Caβ4,HCN3,Kv1.4,Caα1G,Caβ1,
HCN4,Kv3.3and Caβ3;and the ‘PV cluster’ includes
Kvβ1and Caα1A.The biophysical properties ofthe ion
channels in each cluster are consistent with and might
complement each other to generate the three broad
classes of discharge behaviour: ion channels in the
CR cluster are associated with accommodation86;those
in the CB cluster are associated with bursting87;
and those in the PV cluster are associated with high-
Four co-expression principles.These clusters seem to
arise because ofspecific constraints on the types ofgene
that can be co-expressed.The constraints seem to be
governed by four principles:a synergizing principle,
whereby certain gene pairs,such as SK2–CR,that predict
the same electrical phenotype are expressed in the same
cells;an antagonizing principle,whereby certain gene
pairs,such as Kv1.2–Kv3.1and Kv1.2–Kv3.2,that predict
opposite phenotypes are co-expressed;a homogenizing
principle,whereby certain gene pairs,such as Kv1.1–
Kv1.4and PV–Kv1.4,that predict the same phenotype
are expressed in different cells;and a heterogenizing
principle,whereby certain gene pairs,such as HCN4–
PV,Kcn4–Kvβ1,Caβ4–SK2and Caβ4–CR,that predict
opposite phenotypes are expressed in different cells.
These four ion-channel co-expression constraints
could govern the generation of electrical diversity in
Inversion ofexpression.Specific gene-expression profiles
map onto different discharge response types.The most
striking example is a near-perfect inversion of the
expression profile between cells that discharge initially
with a burst onset (b-subtype) and those with a delayed
onset (d-subtype)85.So, even cells that are normally
The gene-expression rules that govern such ‘for-
ward engineering’ ofelectrical behaviour are becoming
clear.The powerful delayed rectifying,voltage-gated
potassium channels Kv3.1 and Kv3.2 are typically
expressed in PV-containing,fast-spiking neocortical
interneurons82–84,although Kv3.1 is also expressed in
PV-negative interneurons and pyramidal neurons.In a
recent study,the mRNA expression of26 ion channels
and 3 calcium-binding proteins was tested in single
The correlation map.This recent study found a signifi-
cant correlation between the ion-channel genes
expressed in an interneuron and its electrical pheno-
type. FIGURE 7 shows the correlation map from this
study,which relates different ion-channel genes to spe-
cific electrical properties (see online supplementary
information S3(table) for electrophysiological parame-
ters).The correlation map provides a coefficient ofcor-
relation for each gene with respect to the value ofeach
electrophysiological parameter.In other words,red pre-
dicts high electrophysiological parameter values ifthe
gene is expressed,and blue predicts low values.The
sum of the colours for all those genes expressed in a
neuron predicts the value ofany ofthe measured elec-
trophysiological parameters (such as the amplitude or
duration ofthe action potential or the rate ofaccom-
modation).The accuracy with which electrophysiologi-
cal parameters can be predicted is surprisingly high,
given the false negatives and the lack of knowledge
about the quantities ofmRNA or protein produced by
each gene,the extent to which each channel is modified
and the distribution patterns ofion channels.The cor-
relation coefficients are also independent ofmorphol-
ogy as they were derived using a training data set that
included neurons from multiple morphological types.
So,merely knowing whether a gene is switched on is
highly informative,and knowing the profile ofexpres-
sion for only a few genes allows the electrical phenotype
ofan interneuron to be predicted.
Table 1 | Electrophysiological classes of neocortical inhibitory neurons
NAC (layers I–VI)
Other classification schemes
AC (layers II–VI)
STUT (layers II–VI)
IS (layers II–V)
BST (layers II–V)
AC, accommodating; b, burst subtype; BSNP, burst spiking non-pyramidal; BST, bursting; c, classical
subtype; d, delay subtype; FS, fast spiking; i, initial; IS, irregular spiking; LS, late spiking; NAC, non-
accommodating; r, repetitive; RSNP, regular spiking non-pyramidal; STUT, stuttering; t, transient.
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |7 9 9
R E V I E W S
Calcium-binding proteins.The three CBPs (CB,PV and
CR) tend to be expressed in three separate populations of
interneurons90,91(see also REFS 8–11,85,89,92),which indi-
cates that an exclusion principle might operate in the
expression of CBPs. However, there is some overlap
in expression,especially between CB and PV11,38and to a
lesser extent between CB and CR11,89.These three popula-
tions correspond approximately to the three broad
discharge response classes:CR in accommodating or
irregular-spiking interneurons77;CB in bursting inter-
neurons74,85;and PV in fast-spiking interneurons74,85,88.
As there are more than three anatomical,electrical and
combined anatomical–electrical types,it is ofcourse not
possible for any one CBP alone to map onto any one type
ofinterneuron (FIG.3).Even the most commonly accepted
notion — that PV is a marker for basket cells — is not
strictly correct,as PV is expressed in only about halfof
basket cells and can also be expressed in ChCs53.The
common use ofPV to isolate fast-spiking cells is also not
entirely correct because not all PV-expressing cells are fast
spiking and not all fast-spiking cells express PV (FIG.3).
Ultimately,the combined expression ofother markers
is required to identify any one of the anatomical–
electrophysiological types ofinterneuron.
Neuropeptides.Other common interneuronal markers
include the neuropeptides SOM,VIP,CCK and NPY93–99.
As with CBPs,no single neuropeptide correlates with
a single anatomical or electrophysiological type of
interneuron11,38,76.However,some expression patterns
are striking,such as SOM expression in MCs100and VIP
expression in SBCs34, DBCs11and BPCs96.Although
some combinations ofneuropeptides map better onto
specific subtypes,they still do not map perfectly9,11,38.
For example, MCs always express SOM and never
VIP11,95(see also REF.56),but this pattern can also be
seen in some BTCs101.Expression patterns,used with
caution,are nevertheless important general indicators
of anatomical and electrophysiological types of
Combined CBP–neuropeptide expression.An inter-
neuron can co-express up to five ofthe seven different
neuropeptides and CBPs38.FIGURE 3shows different types
ofneuron that express neuropeptides and CBPs.At the
protein level,PV,SOM and VIP are found in separate
populations ofneurons11,illustrating another exclusion
principle.However,this exclusion is not perfect at the
mRNA level9,38,85.There might also be an inclusion prin-
ciple,as SOM–NPY11,VIP–CCK102,VIP–CR77and CR–
CCK102co-expression has been detected.Lastly,many of
the markers seem to be expressed independently ofeach
other,providing evidence for an independence principle.
For example, CB is promiscously co-expressed with
many neuropeptides and even other CBPs (see above).
Other markers.Neocortical cells also express distinct
cell-surface molecules (membrane proteins or lipids
with characteristic carbohydrate moieties that can be
identified using antibodies or lectins)10.Futhermore,
pyramidal cells and different types ofinterneuron differ
classified in the same broad class,such as fast spiking,
can have diametrically opposite expression profiles
depending on the onset response.This finding also indi-
cates that only a few transcription factors might control
the expression ofentire sets ofion-channel genes,in
which case it is probable that different combinations of
transcription factors would give rise to a finite number
ofdistinct electrical classes.
Whereas excitatory cells (co-) express only a limited set
of the commonly probed CBPs and neuropeptides,
inhibitory interneurons have more diverse (co-)
expression profiles89(see also REFS 8–11).
c-NAC b-NAC d-NAC
Figure 5 |Different electrophysiological classes of inhibitory interneurons. Five classes
have been observed, based on the steady-state response to a sustained current injection in the
soma: non-accommodating (NAC); accommodating (AC); stuttering (STUT); bursting (BST); and
irregular spiking (IS). Most classes contain three subclasses: delay (d); classic (c) and burst (b). For
bursting interneurons, the three types are repetitive (r), initial (i) and transient (t). RS (regular
spiking) is an example of a classic discharge of a pyramidal cell. See also main text.
8 0 0 |OCTOBER 2004 |VOLUME 5
R E V I E W S
5-methyl-4-isoxazole propionic acid) receptor subunits119,
lack a significant NMDA (N-methyl-D-aspartate) recep-
torcomponent120and often have frequency-dependent
facilitation120–122.These features are found even in some
interneurons with different types ofelectrophysiological
behaviour (classic FS and LTS/BSNP cells)121–123and
indicate that the glutamatergic system might have a
functional dichotomy,with different modes for recruit-
ing pyramidal neurons and interneurons25,120,123,124.
Differential synaptic transmission was confirmed by
simultaneous recordings from a pyramidal neuron that
formed depressing synapses onto other pyramidal
neurons and facilitating synapses onto interneurons125.
Interestingly,the static (quantal) and dynamic (depres-
sion and facilitation) properties offacilitating synapses
from single pyramidal neurons onto interneurons vary
across layers126,which might cause targeted inhibitory
cells in deep cortical layers to discharge before those in
supragranular layers.Such layer-specific differences in
the recruitment of interneurons could influence the
direction ofinformation flow in the cortical column.
Differential synaptic transmission. Although many
connections from pyramidal cells onto interneurons
in neocortical layers II–V show frequency-dependent
facilitation123,125,126,some interneurons receive depressing
synapses from pyramidal neurons121,123,126.Simultaneous
recordings from the same presynaptic pyramidal neuron
and different types ofinterneuron showed that there was
differential glutamatergic transmission onto inter-
neurons126,127; accommodating interneurons with
bitufted dendritic morphologies received facilitating
synapses,whereas non-accommodating interneurons
with multipolar dendritic morphologies (presumably
basket cells) received depressing synapses.Depressing
synapses have also been observed for connections from
pyramidal neurons onto DBCs in layers II/III114(see also
REF.128),onto fast-spiking basket cells in layer V129and
onto irregular-spiking BPCs in layers II/III and V77,indi-
cating that these synapses are more common than was
in their expression ofneurotransmitter receptors103–106
(see also REF.107).Interneurons are therefore diverse in
terms oftheir molecular properties,and the molecular
profile is a tangential dimension that spans interneurons
with different classification rules.
Excitatory synapses on interneurons
Identifying synapses.The first recordings ofsynaptically
connected neurons in the neocortex were performed by
Thomson108.Physiological recordings are the most direct
method for isolating synaptic connections, but it is
important to obtain estimates ofthe numbers and distri-
butions of synapses at the anatomical level. Most
estimates are based on light microscopic analyses and
should be considered estimates ofputative synapses until
verified at the electron microscope level.Nevertheless,
comparisons between light and electron microscope
estimates109–114and between light microscope and
BINOMIAL ESTIMATES15show a reasonable correspondence.
As it is impossible to verify all synapses at the electron
microscope level,light microscopy will remain central in
the study ofneuronal microcircuits.
Anatomical properties.Glutamatergic neurons form mul-
tiple synapses onto interneurons.These synapses typically
form in clusters on a small fraction ofdendrites114,115,in
contrast to the highly distributed innervation ofgluta-
matergic synapses on excitatory cells.Most synapses are
formed on dendrites,but glutamatergic synapses can also
form on the somata ofinterneurons3,116.For example,
pyramidal neurons form about six synapses onto a basket
cell38,117(FIG.4;online supplementary information S2
(table)),with around 60% on ‘basal’dendrites,30% on
the main dendrite and 10% on the soma.There is a corre-
lation between synapse numbers and interneuron types
in the cat,ferret114,118and rat neocortices (FIG.4;online
supplementary information S2(table)).
Physiological properties.Unlike glutamatergic synapses
on excitatory neurons, glutamatergic synapses on
inhibitory cells use different AMPA (α-amino-3-hydroxy-
The number offunctional
release sites is referred to as
binomial nbecause it is
estimated in a quantal analysis
using binomial statistics.
ChCLBCNBCSBC DBCBPC NGCBTC MC
c-ISr-BST b-IS d-STUT b-STUT c-STUTd-NAC c-NAC b-NACb-ACc-ACd-AC
Figure 6 |Anatomical–electrophysiological diversity of neocortical inhibitory neurons. Electrophysiological classification of
interneurons: main classes and subclasses are defined according to discharge responses at steady-state and onset phase to somatic
current injections, respectively. Any given anatomically defined interneuron in general has several distinct discharge behaviours, and,
conversely, a given discharge behaviour can be found in several anatomically defined interneuron types. AC, accommodating;
b, burst subtype; BPC, bipolar cell; BTC, bitufted cell; BST, bursting; c, classic subtype; ChC, chandelier cell; d, delay subtype; DBC,
double bouquet cell; IS, irregular spiking; LBC, large basket cell; MC, Martinotti cell; NAC, non-accommodating; NBC, nest basket
cell; NGC, neurogliaform cell; r, repetitive; SBC, small basket cell; STUT, stuttering.
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |8 0 1
R E V I E W S
the gain of summated potentials and thereby the
action potential discharge of target cells38,50,131.These
interneurons are involved in phasing and synchroniz-
ing neuronal activity132–134.Neurons that preferentially
innervate the dendritic domain are positioned to
influence the dendritic processing and integration of
synaptic inputs135,136;to influence synaptic plasticity
either locally or by interacting with back-propagating
action potentials137;and to affect the generation and
propagation ofdendritic calcium spikes138,139.Finally,
the preferential innervation ofdistal dendritic and tuft
regions could allow neurons to affect local dendritic
Anatomical properties.In general,inhibitory neurons
form more synapses onto their target cells than excita-
tory neurons do (as many as 30 synapses per target,with
an average ofaround 15)14,15,38.Inhibitory synapses are
highly distributed across the dendritic surface oftarget
cells and are mainly formed onto dendritic shafts.
Occasionally,synapses cluster on target cell dendrites112.
The most commonly studied inhibitory neocortical con-
nections are synapses from basket cells onto pyramidal
neurons.Basket cells in layers II–IV form many putative
synapses onto neighbouring pyramidal neurons (LBC:
14.5 ± 1.7 synapses;SBC:20.5 ± 10.5 synapses;NBC:
15.8 ± 4.1;REF.38;FIG.4;online supplementary informa-
tion S2(table)).NBCs in layers II/III and IV did not
show layer-specific differences38,which indicates that the
innervation rules remain consistent across layers.
Synaptic innervation differs for other types ofinterneu-
ron,as well as in different cortical areas,species and
developmental ages112(FIG. 4; online supplementary
Determinants of synaptic dynamics. Although it is
tempting to assume that the physiological properties of
glutamatergic synapses are determined purely by the
postsynaptic target127,128,130,not all multipolar inter-
neurons or FS cells receive depressing synapses,and not
all BTCs,including classic BTCs,receive facilitating
synapses38,123,126.The multipolar–bitufted distinction
has led to many confusing reports and the idea that
dendritic-targeting interneurons receive facilitating
synapses,whereas soma-targeting interneurons receive
depressing synapses,is also mistaken (FIG.8a).Further-
more, a single neocortical neuron can receive both
depressing and facilitating GABA synapses15,indicating
that the target cell does not determine the dynamics of
all incoming synapses. The type of glutamatergic
synapse formed also depends on the electrophysiology
ofthe target cell;ifthe interneuron has a delayed onset
response,synapses tend to be depressing,and if the
onset response is classic or bursting,synapses tend to be
facilitating (A.G.,G.S.and H.M.,unpublished observa-
tions). FIG. 8a shows the mapping of glutamatergic
synapse types onto different interneurons.
Inhibitory synapses on pyramidal neurons
Strategic innervation.Each type ofinterneuron inner-
vates its target cell by preferentially distributing multiple
synapses in a characteristic manner onto selected mem-
brane domains (axon initial segments,somata,proximal
and distal dendritic shafts and spines,and dendritic
tufts)3,14,21.Neurons that preferentially target axon initial
segments are optimally positioned to ‘edit’ a neuron’s
output by affecting the generation and timing ofaction
potentials.The preferential innervation ofthe (peri-)
somatic domain allows presynaptic neurons to control
E61E60 E59E58 E57E56 E55E54 E53 E52E51E50 E49E48E47 E46E45E44 E43E42 E41 E40 E39E38 E37E36E35 E34E33E32 E31E30 E29E28 E27E26E25E24E23 E22 E21E20E19 E18E17E16 E15E14E13 E12E11E10E9 E8E7E6 E5E4E3E2 E1
–1 –0.50 0.51
Figure 7 |Correlation map relating the different ion-channel genes with specific electrical parameters. Ion-channel genes
are indicated on the right, electrical parameters along the top. See online supplementary information S3 (table) for identities of
electrical parameters. The colour indicates the value of the coefficient for each gene, which represents the sign and magnitude of the
correlation between the gene and the value of each electrical parameter. Red indicates that if the gene is expressed, the value of the
electrical parameter will be towards the maximal value recorded in the 203 cells, and vice versa for blue. The value of the electrical
parameter can be calculated by summing the ‘colours’ (coefficients) horizontally for all those genes that were detected in a neuron.
Modified, with permission, from REF.85© (2004) Oxford University Press.
8 0 2 |OCTOBER 2004 |VOLUME 5
R E V I E W S
(NACs,ACs and STUTs,and b-,c- and d-subclasses)
defined interneurons onto pyramidal cells in layers
II–IV15.This study showed that synaptic transmission
mediated by GABAA(GABA type A) receptors was
physiologically much more diverse than previously
reported:inhibitory synapses showed synaptic depres-
sion as well as facilitation to varying degrees,yielding
three distinct classes ofGABA synapse (types I1,I2 and
I3,defined according to the ratio oftime constants of
recovery from synaptic facilitation compared with
depression; FIG.8b).This study also showed that each
type ofinterneuron deploys one ofthese three types of
inhibitory synapse,depending on the anatomical and
physiological properties ofboth pre- and postsynaptic
cells (FIG.8b).This indicates that a ‘pre–post handshake
principle’,the molecular basis of which is unknown,
underlies the formation ofa specific type ofsynapse.
Homogeneous transmission. Remarkably, all the
synapses formed by one interneuron onto multiple
pyramidal neurons show identical synaptic dynamics15.
All the synapses from an interneuron onto all targets of
the same type (pyramidal neurons in this case) seem to
have identical release probabilities and time constants
for recovery from synaptic depression and facilitation.
This homogeneity principle contrasts sharply with the
heterogeneity ofglutamatergic synapses formed by a
pyramidal neuron onto other pyramidal neurons and
also has implications for the forms of learning that
might shape these synapses.The absolute strength of
these synaptic connections is heterogeneous (probably
due to different numbers ofsynapses and/or postsynap-
tic receptors),indicating that they could be modified by
the relative timing ofactivity in only one pair ofneurons
(presynaptic and postsynaptic),but their dynamics are
homogeneous,suggesting that these parameters must be
modified by the activity patterns ofthe entire popula-
tion ofpostsynaptic pyramidal neurons relative to the
single presynaptic interneuron.
GABABtransmission.As well as fast GABAA-receptor-
mediated inhibition,neocortical neurons also show slow
inhibitory synaptic responses mediated by metabotropic
GABAB(GABA type B) receptors.Such responses have
mainly been detected after strong extracellular stimula-
tion or repetitive, high-frequency stimulation of
interneurons140,141.This has led to the idea that the GABAB
receptors are located extrasynaptically and are activated
by GABA ‘spillover’from the synaptic cleft.Alternatively,
neocortical microcircuits might comprise two separate
populations ofinterneurons,each responsible for either
GABAA- or GABAB-receptor-mediated responses142.
Evidence for the segregation ofinterneuron populations
has recently been obtained for a few connections from FS
and RSNP cells onto pyramidal neurons in layer V.
Inhibitory synapses on interneurons
Anatomical properties.With around 50 anatomical–
electrical types ofinterneuron and a handshake principle
for setting synapse types,the number ofpotential types of
connection between inhibitory neurons is enormous.
Physiological properties. An important advance in
understanding the physiological properties ofneocorti-
cal inhibitory circuits came with the systematic study of
connections formed by many anatomically (SBCs,
NBCs,LBCs,MCs and BTCs) and electrophysiologically
Figure 8 |Mapping of synaptic dynamics. Mapping of glutamate (E1 and E2) synapses from
neocortical pyramidal neurons (PCs) onto interneurons (a), and GABA (γ-aminobutyric acid) (I1, I2,
I3) and glutamate (E1and E2) synapses from interneurons onto PCs (b), according to the
anatomical and electrophysiological identity of the different interneurons. Only connections in which
the anatomical and electrical properties were conclusively defined are included. For example, some
b-AC inteneurons generate facilitating input onto PCs (I1) but their anatomical identity has not been
established. In addition, some c-AC interneurons whose morphology was not determined were
also shown to generate facilitating input onto PCs (I1). AC, accommodating; b, burst subtype;
BTC, bitufted cell; c, classic subtype; ChC, chandelier cell; d, delay subtype; DBC, double bouquet
cell; LBC, large basket cell; MC, Martinotti cell; NAC, non-accommodating; NBC, nest basket cell;
NGC, neurogliaform cell; SBC, small basket cell; SSC, spiny stellate cell; STUT, stuttering.
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |8 0 3
R E V I E W S
sufficient sensitivity,complexity and dynamic range for
the inhibitory system to match excitation regardless
of the intensity and complexity of the stimulus, and
synaptic diversity might be crucial to secure the dynamic
range and to choreograph moments of imbalance
between excitation and inhibition in the context ofany
Balancing regions ofneurons.The first challenge is to
balance excitation in different regions ofa single neu-
ron.This might require a range ofinterneurons that are
specialized to target different regions and to collect,
integrate and respond to different types of input.
Balancing excitation in dendrites might require various
interneurons to ‘monitor’excitatory input to many den-
dritic regions, balancing excitation in the cell body
might require interneurons that sense global excitatory
input,and balancing excitation in terms ofthe spike
output might require an interneuron that can collect
sufficient information to veto spiking after integration.
So it is perhaps not so surprising that there is greatest
anatomical variety ofdendritic-targeting interneurons
(DBCs, BTCs, BPs, MCs and NGCs), with a smaller
variety ofsoma-targeting interneurons (LBCs,NBCs
and SBCs) and only one type of axon-taregeting
interneuron (ChCs).Diversity in general,and ofelectri-
cal subtypes in particular,could also be driven by the
need for interneurons to monitor and respond to many
sources of excitatory input (same layer, cross layer,
neighbouring columns, many neocortical regions,
opposite hemisphere and subcortical input).Further
experiments are required to prove that processing input
diversity requires interneuron diversity.
Recruiting balanced inhibition.The second challenge is
for the inhibitory system to sense the appropriate level
ofexcitation across a wide dynamic range and under
various stimulus conditions. A broad spectrum of
action potential thresholds is found in different types
ofinterneuron (thresholds can vary by up to 20 mV;
A.G.,M.T.-R.and H.M.,unpublished observations),
and different discharge rates and patterns might make
this dynamic range possible.The use ofglutamatergic
synapses with varying dynamics could also support the
dynamic range and the sensitivity to specific stimulus
conditions.For example,pyramidal neurons recruit
MCs through facilitating synapses,meaning that during
transient activation ofthe microcircuit,hardly any MCs
will be recruited.However,many LBCs receive depressing
synapses, so they would be instantly recruited; pro-
longed excitation ofthe microcircuit would have the
opposite effect38.Each anatomical–electrophysiological
subtype ofinterneuron therefore has its own conditions
Applying balanced inhibition. Applying the right
amount of inhibition is not a simple process,as only
16% of all synapses on a pyramidal neuron are
inhibitory.The first parameter that needs to be adjusted
to deliver more inhibitory current is the duration of
GABA-receptor activation.The time course ofinhibitory
Nevertheless,there are some general principles.First,it
is rare to record connections between interneurons,
indicating that they are sparsely connected15,38,143.
Second, connections between interneurons are also
mediated by multiple synapses (an average ofaround
ten).Third,interneurons can target specific domains of
other interneurons.Last,inhibitory synapses onto inter-
neurons are highly distributed on the dendrites,as for
pyramidal neurons.Only a few ofthe possible combina-
tions ofinterneuron–interneuron synaptic connection
have been studied,possibly because specific types of
interneuron are preferentially interconnected.Basket
cells innervate neighbouring basket cells,dendritic-
targeting cells and DBCs113.These synapses are located
somatically and peri-somatically, but whereas both
dendritic-targeting cells and DBCs receive only a few
synapses,postsynaptic basket cells receive more (FIG.4;
online supplementary information S2(table)).Such
differences have been interpreted to indicate both
selectivity and preference in inhibitory neocortical
microcircuits14.The large numbers ofsynapses between
basket cells113might underlie extensive basket-cell
networks,which have been implicated in long-range
lateral disinhibition144.Activity in such networks might
also be coordinated by electrical synapses145–149.
Physiological properties. Only a few types of inter-
neuron–interneuron connection have been studied
physiologically.Fast-spiking basket cells in layers II–IV
form depressing synapses onto other basket cells,DBCs
and dendritic-targeting cells113.Divergent connections
onto non-accommodating multipolar interneurons (pos-
sibly basket cells) and accommodating interneurons with
bitufted dendrites show different degrees ofdepression127.
Interneurons can deploy any of the three types of
synapse15(FIG.8b).Although these studies seem to imply
that each type of interneuron can innervate, and be
innervated by,all other types,this is not likely.Inter-
interneuron connections would need to be orders of
magnitude greater for ‘all-types to all-types’connectivity.
Indeed,some interneurons,such as ChCs,do not contact
certain other types ofinterneuron at all21.Finally,there is
anatomical and physiological evidence that some
interneuron types, such as LBCs, are highly inter-
connected,whereas other types,such as DBCs,are much
Function of interneuron diversity
Sensory stimulation results in a coordinated increase in
excitatory and inhibitory conductances151,152.Surprisingly,
a balance between these two opposing conductances is
maintained over a large dynamic range and for many
stimuli152,153.This means that regardless ofthe level of
excitation,the inhibitory system can automatically scale
its output to provide matching opposition across a large
dynamic range, analogous to a Yin-like inhibition
opposing a Yang-like excitation154. Excitatory and
inhibitory inputs are,however,distributed in various
temporal combinations,and large imbalances can occur
transiently during the response to stimuli151–153,155–157.
Interneuron diversity might be crucial for providing
8 0 4 |OCTOBER 2004 |VOLUME 5
R E V I E W S
and H.M.,unpublished observations;see also REF.92).
At the electrical level,the diversity might seem arbitrary,
but this is probably due to the lack ofdefined functions
for the different behaviours.The class issue at all levels
will probably only be resolved objectively at the level of
gene expression.The correlation between expression
profiles and electrical phenotypes,the constraints in
co-expression profiles and the ‘flip’ ofentire expression
profiles to form opposite electrical phenotypes85all
indicate that only a few transcription factors,expressed
in different combinations, might give rise to a finite
number of distinct classes of interneuron. So, most
interneurons probably lie in distinct electrical,morpho-
logical and molecular classes.The observed diversity is
several orders ofmagnitude smaller than expected for
a continuum of electrical types using more than 100
ion-channel genes,indicating powerful constraints on
diversity.Understanding these constraints is also key to
resolving the class-versus-continuum debate.
Stereotypical GABA innervation
The neocortical microcircuitry is stereotypical in many
respects166,but subtle variations become apparent as the
microcircuit is studied at higher resolution.For example,
inhibitory input is stereotypical in that all pyramidal cells
receive inputs from the three broad types ofinterneuron
(dendritic,somatic and axon-targeting),but does each
pyramidal neuron receive inputs from the same subtypes
ofinterneuron? About 16% ofthe synapses on a pyrami-
dal neuron in layers II/III are inhibitory1,18,and these
come from about 70 interneurons (half being basket
cells).Roughly 50 ofthese interneurons arise from the
same layer and column as the pyramidal neuron, 10
from the same column but a different layer,and 10 from
different columns (A.G.and H.M.,unpublished observa-
tions).Each of the 70 interneurons places around 15
synapses on the pyramidal neuron, together making
around 1,000 inhibitory synapses.In principle,therefore,
each pyramidal cell could receive input from at least one
ofeach anatomical type ofinterneuron needed to inner-
vate all parts ofthe neuron.However,it is unlikely that all
the different electrical subtypes could be represented in
their correct proportions on every pyramidal neuron.
If anatomical–electrical–molecular variants are also
considered,then it is certainly not possible for each pyra-
midal neuron to receive inputs from an identical set of
interneurons.The next question is whether there is such
high-resolution stereotypy for small ‘sets’ ofpyramidal
neurons,where sets receive inputs from the same com-
plement ofanatomical–electrical–molecular subtypes of
interneuron.This might be possible within a layer,but as
there are layer-specific differences in molecular expres-
sion profiles and electrical subtypes ofinterneuron,such
high-resolution stereotypy will not hold across layers.
The fundamental question now is how microcircuits in
different species, different brain regions of the same
species,different layers and even different neurons in the
same layer are driven to diversify to form countless varia-
tions of the microcircuit template — in particular,
whether stimulus diversity is the ultimate driving force
behind interneuron diversity.
synaptic currents is about twofold longer than for excita-
tory currents15,159. This is still not enough to deliver
matching inhibition across a large dynamic range,so
interneurons can discharge 2–3 times faster than pyra-
midal neurons. To make use of these properties,
inhibitory synapses have greater synaptic facilitation
than excitatory synapses,which allows transmission at
Balancing the circuit.The fourth challenge in the balanc-
ing act is to apply inhibition at the right moment in each
neuron ofthe microcircuit.In particular,the timing and
amount ofinhibition to each pyramidal neuron — sub-
sets ofwhich receive input from and transmit output to
different locations — must be orchestrated in a stimulus-
dependent manner.Dynamic synapses choreograph pre-
cise millisecond timing ofsynaptic activity in different
interneurons relative to pyramidal neurons in a manner
that depends on the structure ofthe stimulus158.
Why balance Yang with Yin?Why does excitation need to
be balanced with inhibition and why do transient
moments ofimbalance occur? This is a vast area,which
will not be dealt with in this review,except to speculate on
two potential reasons.At the level ofindividual neurons,
matching inhibition as a function ofstimulus intensity
could allow information to be processed and encoded at a
higher or lower temporal resolution,depending on the
baseline firing rates.This can be achieved by changing
the membrane time constant,which changes the time
window for temporal integration161(see also REF.162) and
by changing the temporal precision ofspike generation by
adding high-frequency membrane ‘noise’163–165.At this
level,balance might be required to normalize the baseline
for synaptic integration as a function ofactivity (to nor-
malize the mutual information between the input chan-
nels) and spiking might reflect moments ofimbalance
(high mutual information between the input channels).
At the microcircuit level,a sliding scale between integra-
tion and coincidence detection as a function ofactivity in
each neuron could be important to control which neu-
rons synchronize at which frequencies162.Balance might
be required to keep all neurons independent (to normal-
ize mutual information across neurons) and oscillations
might reflect orchestrated momentary imbalances of
groups ofneurons (high mutual information between
neurons).Needless to say,considerable work is required
to test and turn theory into fact.
Classes or a continuum?
At the anatomical level,neocortical interneurons are gen-
erally accepted as being in distinct classes,not because of
any objective analyses,but because ofmore obvious func-
tional specializations indicated by their different domain-
targeting tendencies.At the molecular level,the issue is,
on the one hand, simpler because some markers are
expressed only by certain interneuron types,but,on the
other hand,more complex because no one marker points
unambiguously towards only one anatomical or electrical
type ofinterneuron;the expression pattern offour or five
markers might be required (M.T.-R.,M.Ilic,P.Goodman
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |8 0 5
R E V I E W S
support simultaneous processing ofmultiple features,
and has immense flexibility to dynamically re-configure,
forming a transient task-specific microcircuit. The
diversity ofinhibitory interneurons and synapses could
be crucial to impart this ‘omnipotence’ to a micro-
circuit,which is required to support optimal informa-
tion processing ofany stimulus in any species and brain
region in a rapidly changing and unpredictable
A template neocortical microcircuit seems to have
been duplicated repeatedly to construct the neocortex.
In terms of its general architecture,this template is
highly stereotypical — a ‘generic’ microcircuit — with
subtle specializations that presumably optimize pro-
cessing in different brain regions and species to form a
‘task-specific’ microcircuit.In terms of its function,
this generic microcircuit at any one location seems to
1. DeFelipe, J . & Farinas, I. The pyramidal neuron of the
cerebral cortex: morphological and chemical characteristics
of the synaptic inputs. Prog. Neurobiol. 39, 563–607
J ones, E. G. in Cellular Components of the Cerebral Cortex
(eds Peters, A. & J ones, E. G.) 521–554 (Plenum, New York,
White, E. L. Cortical Circuits. Synaptic Organization of the
Cerebral Cortex (Birkhauser, Boston, 1989).
Ren, J . Q., Aika, Y., Heizmann, C. W. & Kosaka, T.
Quantitative analysis of neurons and glial cells in the rat
somatosensory cortex, with special reference to GABAergic
neurons and parvalbumin-containing neurons. Exp. Brain
Res. 92, 1–14 (1992).
Careful estimates of neuron and synapse numbers.
Beaulieu, C. Numerical data on neocortical neurons in adult
rat, with special reference to the GABA population. Brain
Res. 609, 284–292 (1993).
Peters, A. & J ones, E. G. (eds) Cellular Components of the
Cerebral Cortex (Plenum, New York, 1984).
Peters, A. & Sethares, C. Organization of pyramidal neurons
in area 17 of monkey visual cortex. J . Comp. Neurol. 306,
Toledo-Rodriguez, M., Gupta, A., Wang, Y., Wu, C. Z. &
Markram, H. in The Handbook of Brain Theory and Neural
Networks. (ed. Arbib, M. A.) 719–725 (MIT Press,
Cambridge, Massachusetts., 2003).
Cauli, B. et al. Molecular and physiological diversity of
cortical non pyramidal cells. J . Neurosci. 17, 3894–3906
10. DeFelipe, J . Neocortical neuronal diversity: chemical
heterogeneity revealed by co-localization studies of classic
neurotransmitters, neuropeptides, calcium-binding proteins,
and cell surface molecules. Cereb. Cortex3, 273–289
A key review that reveals different chemical
subpopulations of neocortical neurons.
11. Kawaguchi, Y. & Kubota, Y. GABAergic cell subtypes and
their synaptic connections in rat frontal cortex. Cereb.
Cortex7, 476–486 (1997).
A multi-dimensional study that emphasizes
characteristic physiological, morphological and
immunohistochemical features of some non-
pyramidal cells, also in combination with synaptic
innervation properties. It also shows that vasoactive
intestinal peptide, parvalbumin and somatostatin are
expressed by distinct subpopulations of interneurons.
12. DeFelipe, J . Cortical interneurons: from Cajal to 2001. Prog.
Brain Res. 136, 215–238 (2002).
An excellent review on the history of discovery of
different cortical neurons.
13. Houser, C. R., Vaughn, J . E., Hendry, S. H., J ones, E. G. &
Peters, A. in Cerebral Cortex: Functional Properties of
Cortical Cells (eds J ones, E. G. & Peters, A.) 63–90 (Plenum,
New York, 1984).
14. Somogyi, P., Tamas, G., Lujan, R. & Buhl, E. H. Salient
features of synaptic organisation in the cerebral cortex. Brain
Res. Brain Res. Rev. 26, 113–135 (1998).
15. Gupta, A., Wang, Y. & Markram, H. Organizing principles for
a diversity of GABAergic interneurons and synapses in the
neocortex. Science287, 273–278 (2000).
The first study to report three distinct types of GABA
synapse and several principles that determine which
type of synapse is deployed by different anatomical
and electrophysiological types of interneuron when
inhibitory connections are formed.
16. Thomson, A. M. & Deuchars, J . Temporal and spatial
properties of local circuits in neocortex. Trends Neurosci.
17, 119–126 (1994).
17. Douglas, R. & Martin, K. A. in The Synaptic Organization of
the Brain459–511(Oxford Univ. Press, New York, 1998).
18. Peters, A. in Synaptic Functions (eds Edelman, G. M., Gall,
W. E. & Cowan, W. M.) 373–397 (Wiley, New York, 1987).
19. Fairen, A., DeFelipe, J . & Regidor, J . in Cellular Components
of the Cerebral Cortex(eds Peters, A. & J ones, E. G.)
206–241 (Plenum, New York, 1984).
20. Letinic, K., Zoncu, R. & Rakic, P. Origin of GABAergic
neurons in the human neocortex. Nature417, 645–649
21. DeFelipe, J . Types of neurons, synaptic connections and
chemical characteristics of cells immunoreactive for
calbindin-D28K, parvalbumin and calretinin in the neocortex.
J . Chem. Neuroanat. 14, 1–19 (1997).
22. Lund, J . S. Organization of neurons in the visual cortex, area
17, of the monkey (Macaca mulatta). J . Comp. Neurol. 147,
23. LeVay, S. Synaptic patterns in the visual cortex of the cat
and monkey. Electron microscopy of Golgi preparations.
J . Comp. Neurol. 150, 53–85 (1973).
24. Feldmeyer, D., Lubke, J ., Silver, R. A. & Sakmann, B.
Synaptic connections between layer 4 spiny neuron-layer
2/3 pyramidal cell pairs in juvenile rat barrel cortex:
physiology and anatomy of interlaminar signalling within a
cortical column. J . Physiol. (Lond.)538, 803–822 (2002).
The first study to reveal a unidirectional pathway from
spiny stellate cells in layer IV to layer II/III pyramidal
25. Thomson, A. M. Activity-dependent properties of synaptic
transmission at two classes of connections made by rat
neocortical pyramidal axons in vitro. J . Physiol. (Lond.)502,
26. Thomson, A. M., West, D. C., Wang, Y. & Bannister, A. P.
Synaptic connections and small circuits involving excitatory
and inhibitory neurons in layers 2–5 of adult rat and cat
neocortex: triple intracellular recordings and biocytin
labelling in vitro. Cereb. Cortex12, 936–953 (2002).
27. Lund, J . S. in Cellular Components of the Cerebral Cortex
(eds Peters, A. & J ones, E. G.) 255–308 (Plenum, New York,
28. Cajal, S. R. Histology Due Systeme Nerveux de Homme et
des Vertebrates (Maloine, Paris, 1909).
29. Callaway, E. M. Local circuits in primary visual cortex of the
macaque monkey. Annu. Rev. Neurosci. 21, 47–74 (1998).
30. J ones, E. G. Varieties and distribution of non-pyramidal cells
in the somatic sensory cortex of the squirrel monkey.
J . Comp. Neurol. 160, 205–267 (1975).
31. Gilbert, C. D. Circuitry, architecture, and functional dynamics
of visual cortex. Cereb. Cortex3, 373–386 (1993).
32. Keller, A. Intrinsic synaptic organization of the motor cortex.
Cereb. Cortex3, 430–441 (1993).
33. Lund, J. S. Anatomical organization of macaque monkey
striate visual cortex. Annu. Rev. Neurosci.11, 253–288 (1988).
34. Valverde, F. Intrinsic neocortical organization: some
comparative aspects. Neuroscience18, 1–23 (1986).
35. Feldman, M. L. & Peters, A. The forms of non-pyramidal
neurons in the visual cortex of the rat. J . Comp. Neurol.
179, 761–793 (1978).
36. Martin, K. A. & Whitteridge, D. Form, function and
intracortical projections of spiny neurons in the striate visual
cortex of the cat. J . Physiol. (Lond.)353, 463–504 (1984).
37. Peters, A. & Fairen, A. Smooth and sparsely-spined stellate
cells in the visual cortex of the rat: a study using a combined
Golgi-electron microscopic technique. J . Comp. Neurol.
181, 129–171 (1978).
38. Wang, Y., Gupta, A., Toledo-Rodriguez, M., Wu, C. Z. &
Markram, H. Anatomical, physiological, molecular and
circuit properties of nest basket cells in the developing
somatosensory cortex. Cereb. Cortex12, 395–410 (2002).
The first study to define the nest basket cell as a
distinct subclass of basket cell.
39. Somogyi, P. in Neuronal Mechanisms of Visual Perception,
Proc. Retina Res. Found. Symp. 2 (eds Lamm, D. K. &
Gilbert, C. D.) 35–62 (Portfolio, Woodlands, Texas, 1989).
40. Marin-Padilla, M. Origin of the pericellular baskets of the
pyramidal cells of the human motor cortex: a Golgi study.
Brain Res. 14, 633–646 (1969).
41. Kisvarday, Z. F. & Eysel, U. T. Cellular organization of
reciprocal patchy networks in layer III of cat visual cortex
(area 17). Neuroscience46, 275–286 (1992).
42. Szentagothai, J. in Central Processing of Visual Information, B.
Visual Centers in the Brain(ed. J ung, R.) 269–324 (Springer,
43. Kisvarday, Z. F., Martin, K. A., Whitteridge, D. & Somogyi, P.
Synaptic connections of intracellularly filled clutch cells: a
type of small basket cell in the visual cortex of the cat.
J . Comp. Neurol. 241, 111–137 (1985).
44. DeFelipe, J . & Fairen, A. A type of basket cell in superficial
layers of the cat visual cortex. A Golgi-electron microscope
study. Brain Res. 244, 9–16 (1982).
45. Peters, A. & J ones, E. G. in Cellular Components of the
Cerebral Cortex(eds Peters, A. & J ones, E. G.) 107–122
(Plenum, New York, 1984).
46. Lund, J . S. & Lewis, D. A. Local circuit neurons of
developing and mature macaque prefrontal cortex: Golgi
and immunocytochemical characteristics. J . Comp. Neurol.
328, 282–312 (1993).
47. Somogyi, P. A specific ‘axo-axonal’ interneuron in the visual
cortex of the rat. Brain Res. 136, 345–350 (1977).
This paper describes the seminal discovery of the
48. Fairen, A. & Valverde, F. A specialized type of neuron in the
visual cortex of cat: a Golgi and electron microscope study
of chandelier cells. J . Comp. Neurol. 194, 761–779 (1980).
49. Buhl, E. H., Halasy, K. & Somogyi, P. Diverse sources of
hippocampal unitary inhibitory postsynaptic potentials and the
number of synaptic release sites. Nature368, 823–828 (1994).
50. Miles, R., Toth, K., Gulyas, A. I., Hajos, N. & Freund, T. F.
Differences between somatic and dendritic inhibition in the
hippocampus. Neuron16, 815–823 (1996).
Important differences in the functional impact of
somatic- and dendritic-targeting interneurons were
revealed by this study.
51. Zhu, Y., Stornetta, R. L. & Zhu, J . J . Chandelier cells control
excessive cortical excitation: characteristics of whisker-
evoked synaptic responses of layer 2/3 nonpyramidal and
pyramidal neurons. J . Neurosci. 24, 5101–5108 (2004).
52. Somogyi, P., Freund, T. F. & Cowey, A. The axo-axonic
interneuron in the cerebral cortex of the rat, cat and monkey.
Neuroscience7, 2577–2607 (1982).
53. DeFelipe, J . Chandelier cells and epilepsy. Brain122,
54. DeFelipe, J ., Hendry, S. H. & J ones, E. G. Visualization of
chandelier cell axons by parvalbumin immunoreactivity in
monkey cerebral cortex. Proc. Natl Acad. Sci. USA 86,
55. Ganter, P., Szucs, P., Paulsen, O. & Somogyi, P. Properties
of horizontal axo-axonic cells in stratum oriens of the
hippocampal CA1 area of rats in vitro. Hippocampus 14,
56. Wang, Y. et al. Anatomical, physiological and molecular
properties of Martinotti cells in the somatosensory cortex of
the juvenile rat. J . Physiol. (Lond.)26 Aug 2004 [epub ahead
57. Braitenberg, V. & Schüz, A. Cortex: Statistics and Geometry
of Neural Connectivity(Springer, Heidelberg, 1998).
58. Peters, A. in Cellular Components of the Cerebral Cortex(eds
Peters, A. & J ones, E. G.) 381–408 (Plenum, New York, 1984).
59. Peters, A. & Harriman, K. M. Enigmatic bipolar cell of rat
visual cortex. J . Comp. Neurol. 267, 409–432 (1988).
60. Peters, A. The axon terminals of vasoactive intestinal
polypeptide (VIP)-containing bipolar cells in rat visual cortex.
J . Neurocytol. 19, 672–685 (1990).
61. Somogyi, P. & Cowey, A. Combined Golgi and electron
microscopic study on the synapses formed by double
bouquet cells in the visual cortex of the cat and monkey.
J . Comp. Neurol. 195, 547–166 (1981).
62. Somogyi, P. & Cowey, A. in Cellular Components of the
Cerebral Cortex(eds Peters, A. & J ones, E. G.) 337–360
(Plenum, New York, 1984).
8 0 6 |OCTOBER 2004 |VOLUME 5
R E V I E W S
63. DeFelipe, J ., Hendry, S. H., Hashikawa, T., Molinari, M. &
J ones, E. G. A microcolumnar structure of monkey cerebral
cortex revealed by immunocytochemical studies of double
bouquet cell axons. Neuroscience37, 655–673 (1990).
64. J ones, E. G. in Cellular Components of the Cerebral Cortex
(eds Peters, A. & J ones, E. G.) 409–418 (Plenum, New York,
65. Marin-Padilla, M. in Cellular Components of the Cerebral
Cortex(eds Peters, A. & J ones, E. G.) 447–478 (Plenum,
New York, 1984).
66. Hestrin, S. & Armstrong, W. E. Morphology and physiology
of cortical neurons in layer I. J . Neurosci. 16, 5290–5300
67. Anderson, J . C., Martin, K. A. & Picanco-Diniz, C. W. The
neurons in layer 1 of cat visual cortex. Proc. R. Soc. Lond. B
248, 27–33 (1992).
68. Zhou, F. M. & Hablitz, J . J . Morphological properties of
intracellularly labeled layer I neurons in rat neocortex.
J . Comp. Neurol. 376, 198–213 (1996).
69. McCormick, D. A., Connors, B. W., Lighthall, J . W. &
Prince, D. A. Comparative electrophysiology of pyramidal
and sparsely spiny stellate neurons of the neocortex.
J . Neurophysiol. 54, 782–806 (1985).
The pioneering study that revealed basic differences
in discharge between pyramidal neurons and
70. Gutnick, M. J . & Crill, W. E. in The Cortical Neuron(eds
Gutnick, M. J . & Mody, I.) 33–51 (Oxford Univ. Press, New
71. Connors, B. W. & Gutnick, M. J . Intrinsic firing patterns of
diverse neocortical neurons. Trends Neurosci. 13, 99–104
72. Amitai, Y. & Connors, B. W. in Cerebral Cortex(eds Jones, E. G.
& Diamond, I. T.) 299–331 (Plenum, New York, 1995).
73. Kawaguchi, Y. Groupings of nonpyramidal and pyramidal
cells with specific physiological and morphological
characteristics in rat frontal cortex. J . Neurophysiol.
69, 416–431 (1993).
74. Kawaguchi, Y. & Kubota, Y. Correlation of physiological
subgroupings of nonpyramidal cells with parvalbumin- and
calbindinD28k-immunoreactive neurons in layer V of rat
frontal cortex. J . Neurophysiol. 70, 387–396 (1993).
75. Kawaguchi, Y. & Kubota, Y. Physiological and morphological
identification of somatostatin- or vasoactive intestinal
polypeptide-containing cells among GABAergic cell
subtypes in rat frontal cortex. J . Neurosci. 16, 2701–2715
76. Kawaguchi, Y. & Kubota, Y. Neurochemical features and
synaptic connections of large physiologically-identified
GABAergic cells in the rat frontal cortex. Neuroscience85,
77. Porter, J . T. et al. Properties of bipolar VIPergic interneurons
and their excitation by pyramidal neurons in the rat
neocortex. Eur. J . Neurosci. 10, 3617–3628 (1998).
78. Kawaguchi, Y. Physiological subgroups of nonpyramidal
cells with specific morphological characteristics in layer II/III
of rat frontal cortex. J . Neurosci. 15, 2638–2655 (1995).
79. Steriade, M. Corticothalamic resonance, states of vigilance
and mentation. Neuroscience101, 243–276 (2000).
80. Llinas, R. The intrinsic electrophysiological properties of
mammalian neurons: insights into central nervous system
function. Science242, 1654–1664 (1988).
81. Mainen, Z. F. & Sejnowski, T. J . Influence of dendritic
structure on firing pattern in model neocortical neurons.
Nature382, 363–366 (1996).
82. Rudy, B. & McBain, C. J . Kv3 channels: voltage-gated K+
channels designed for high-frequency repetitive firing.
Trends Neurosci. 24, 517–526 (2001).
83. Martina, M., Schultz, J. H., Ehmke, H., Monyer , H. & Jonas, P.
Functional and molecular differences between voltage-
gated K+channels of fast-spiking interneurons and
pyramidal neurons of rat hippocampus. J . Neurosci. 18,
The first combined patch-clamp reverse transcription
PCR study, showing differences in ion channels in
pyramidal neurons and interneurons.
84. Erisir, A., Lau, D., Rudy, B. & Leonard, C. S. Function of
specific K+channels in sustained high-frequency firing of
fast-spiking neocortical interneurons. J . Neurophysiol. 82,
85. Toledo-Rodriguez, M. et al. Correlation maps allow neuronal
electrical properties to be predicted from single-cell gene
expression profiles in rat neocortex. Cereb. Cortex10 J une
2004 [epub ahead of print].
The first study to reveal profiles of ion-channel and
calcium-binding protein genes expressed in
neocortical neurons and to use expression profiles to
predict electrical behaviour.
86. Vergara, C., Latorre, R., Marrion, N. V. & Adelman, J . P.
Calcium-activated potassium channels. Curr. Opin.
Neurobiol. 8, 321–329 (1998).
87. Ertel, S. & Ertel, E. Low-voltage-activated T-type Ca2+
channels. Trends Pharmacol. Sci. 18, 37–42 (1997).
88. Chow, A. et al. K+channel expression distinguishes
subpopulations of parvalbumin- and somatostatin-
containing neocortical interneurons. J . Neurosci. 19,
89. Kubota, Y., Hattori, R. & Yui, Y. Three distinct
subpopulations of GABAergic neurons in rat frontal
agranular cortex. Brain Res. 649, 159–173 (1994).
A seminal study showing that three distinct
subpopulations of interneurons express parvalbumin,
calretinin or somatostatin.
90. Demeulemeester, H., Vandesande, F., Orban, G. A.,
Heizmann, C. W. & Pochet, R. Calbindin D-28K and
parvalbumin immunoreactivity is confined to two separate
neuronal subpopulations in the cat visual cortex, whereas
partial coexistence is shown in the dorsal lateral geniculate
nucleus. Neurosci. Lett. 99, 6–11 (1989).
91. Rogers, J . H. & Resibois, A. Calretinin and calbindin-D28k in
rat brain: patterns of partial co-localization. Neuroscience
51, 843–865 (1992).
References 90 and 91 revealed the differential
expression of calcium-binding proteins in GABA
92. Cauli, B. et al. Classification of fusiform neocortical
interneurons based on unsupervised clustering. Proc. Natl
Acad. Sci. USA 97, 6144–6149 (2000).
93. Hendry, S. H., J ones, E. G. & Emson, P. C. Morphology,
distribution, and synaptic relations of somatostatin- and
neuropeptide Y-immunoreactive neurons in rat and monkey
neocortex. J . Neurosci. 4, 2497–2517 (1984).
94. Rogers, J . H. Immunohistochemical markers in rat cortex:
co-localization of calretinin and calbindin-D28k with
neuropeptides and GABA. Brain Res. 587, 147–157
95. Demeulemeester, H., Vandesande, F., Orban, G. A.,
Brandon, C. & Vanderhaeghen, J . J . Heterogeneity of
GABAergic cells in cat visual cortex. J . Neurosci. 8,
96. Morrison, J . H., Magistretti, P. J ., Benoit, R. & Bloom, F. E.
The distribution and morphological characteristics of the
intracortical VIP-positive cell: an immunohistochemical
analysis. Brain Res. 292, 269–282 (1984).
97. J ones, E. G. & Hendry, S. H. Peptide-containing neurons of
the primate cerebral cortex. Res. Publ. Assoc. Res. Nerv.
Ment. Dis. 64, 163–178 (1986).
98. Somogyi, P. et al. Different populations of GABAergic
neurons in the visual cortex and hippocampus of cat contain
somatostatin- or cholecystokinin-immunoreactive material.
J . Neurosci. 4, 2590–2603 (1984).
99. Hendry, S. H. et al. Neuropeptide-containing neurons of the
cerebral cortex are also GABAergic. Proc. Natl Acad. Sci.
USA 81, 6526–6530 (1984).
References 98 and 99 reported the differential
expression of intestinal peptides in GABA neurons.
100. Wahle, P. Differential regulation of substance P and
somatostatin in Martinotti cells of the developing cat visual
cortex. J . Comp. Neurol. 329, 519–538 (1993).
101. Meinecke, D. L. & Peters, A. Somatostatin immunoreactive
neurons in rat visual cortex: a light and electron microscopic
study. J . Neurocytol. 15, 121–136 (1986).
102. Kubota, Y. & Kawaguchi, Y. Two distinct subgroups of
cholecystokinin-immunoreactive cortical interneurons. Brain
Res. 752, 175–183 (1997).
103. J onas, P., Racca, C., Sakmann, B., Seeburg, P. H. &
Monyer, H. Differences in Ca2+permeability of AMPA-type
glutamate receptor channels in neocortical neurons caused
by differential GluR-B subunit expression. Neuron12,
104. Stewart, A. E., Yan, Z., Surmeier, D. J . & Foehring, R. C.
Muscarine modulates Ca2+channel currents in rat
sensorimotor pyramidal cells via two distinct pathways.
J . Neurophysiol. 81, 72–84 (1999).
105. Angulo, M. C., Lambolez, B., Audinat, E., Hestrin, S. &
Rossier, J . Subunit composition, kinetic, and permeation
properties of AMPA receptors in single neocortical
nonpyramidal cells. J . Neurosci. 17, 6685–6696 (1997).
106. Flint, A. C., Maisch, U. S., Weishaupt, J . H., Kriegstein, A. R.
& Monyer, H. NR2A subunit expression shortens NMDA
receptor synaptic currents in developing neocortex.
J . Neurosci. 17, 2469–2476 (1997).
107. Monyer, H. & Markram, H. Interneuron diversity series:
molecular and genetic tools to study GABAergic
interneuron diversity and function. Trends Neurosci. 27,
108. Thomson, A. M., Girdlestone, D. & West, D. C. Voltage-
dependent currents prolong single-axon postsynaptic
potentials in layer III pyramidal neurons in rat neocortical
slices. J . Neurophysiol. 60, 1896–1907 (1988).
This paper reported the first dual recordings of
synaptically connected neurons in the neocortex.
109. Lubke, J ., Markram, H., Frotscher, M. & Sakmann, B.
Frequency and dendritic distribution of autapses established
by layer 5 pyramidal neurons in the developing rat neocortex:
comparison with synaptic innervation of adjacent neurons of
the same class. J . Neurosci. 16, 3209–3218 (1996).
110. Markram, H., Lubke, J., Frotscher , M., Roth, A. & Sakmann, B.
Physiology and anatomy of synaptic connections between
thick tufted pyramidal neurons in the developing rat
neocortex. J . Physiol. (Lond.)500, 409–440 (1997).
111. Silver, R. A., Lubke, J ., Sakmann, B. & Feldmeyer, D. High-
probability uniquantal transmission at excitatory synapses in
barrel cortex. Science302, 1981–1984 (2003).
112. Tamas, G., Buhl, E. H. & Somogyi, P. Fast IPSPs elicited via
multiple synaptic release sites by different types of
GABAergic neuron in the cat visual cortex. J . Physiol.
(Lond.) 500, 715–738 (1997).
113. Tamas, G., Somogyi, P. & Buhl, E. H. Differentially
interconnected networks of GABAergic interneurons in the
visual cortex of the cat. J . Neurosci. 18, 4255–4270 (1998).
114. Buhl, E. H. et al. Effect, number and location of synapses
made by single pyramidal cells onto aspiny interneurons of
cat visual cortex. J . Physiol. (Lond.)500, 689–713 (1997).
An elegant multi-dimensional study of a key
115. Ahmed, B., Anderson, J . C., Martin, K. A. & Nelson, J . C.
Map of the synapses onto layer 4 basket cells of the primary
visual cortex of the cat. J . Comp. Neurol. 380, 230–242
116. Peters, A., Palay, S. L. & Webster, H. D. The Fine Structure of
the Nervous System (Oxford Univ. Press, New York, 1991).
117. Deuchars, J . & Thomson, A. M. Innervation of burst firing
spiny interneurons by pyramidal cells in deep layers of rat
somatomotor cortex: paired intracellular recordings with
biocytin filling. Neuroscience69, 739–755 (1995).
118. Krimer, L. S. & Goldman-Rakic, P. S. Prefrontal microcircuits:
membrane properties and excitatory input of local, medium,
and wide arbor interneurons. J . Neurosci. 21, 3788–3796
119. Hestrin, S. Different glutamate receptor channels mediate
fast excitatory synaptic currents in inhibitory and excitatory
cortical neurons. Neuron11, 1083–1091 (1993).
120. Thomson, A. M., Deuchars, J . & West, D. C. Neocortical
local synaptic circuitry revealed with dual intracellular
recordings and biocytin-filling. J . Physiol. (Paris)90,
121. Thomson, A. M., West, D. C. & Deuchars, J . Properties of
single axon excitatory postsynaptic potentials elicited in
spiny interneurons by action potentials in pyramidal neurons
in slices of rat neocortex. Neuroscience69, 727–738 (1995).
122. Thomson, A. M., Deuchars, J . & West, D. C. Single axon
excitatory postsynaptic potentials in neocortical
interneurons exhibit pronounced paired pulse facilitation.
Neuroscience54, 347–360 (1993).
The first demonstration of strongly facilitating
123. Thomson, A. M. & Deuchars, J . Synaptic interactions in
neocortical local circuits: dual intracellular recordings in vitro.
Cereb. Cortex7, 510–522 (1997).
124. Thomson, A. M. Neuroscience. More than just frequency
detectors? Science275, 179–180 (1997).
125. Markram, H., Wang, Y. & Tsodyks, M. Differential signaling
via the same axon of neocortical pyramidal neurons. Proc.
Natl Acad. Sci. USA 95, 5323–5328 (1998).
The first direct demonstration that the same axon
from a neocortical neuron can form both depressing
and facilitating synapses.
126. Wang, Y., Gupta, A. & Markram, H. Anatomical and
functional differentiation of glutamatergic synaptic
innervation in the neocortex. J . Physiol. (Paris)93, 305–317
127. Reyes, A. et al. Target-cell-specific facilitation and
depression in neocortical circuits. Nature Neurosci.
1, 279–285 (1998).
128. Rozov, A., J erecic, J ., Sakmann, B. & Burnashev, N.
AMPA receptor channels with long-lasting desensitization in
bipolar interneurons contribute to synaptic depression in a
novel feedback circuit in layer 2/3 of rat neocortex. J. Neurosci.
21, 8062–8071 (2001).
129. Galaretta, M. & Hestrin, S. Frequency-dependent synaptic
depression and the balance of excitation and inhibition in the
neocortex. Nature Neurosci. 1, 587–594 (1998).
130. Rozov, A., Burnashev, N., Sakmann, B. & Neher, E.
Transmitter release modulation by intracellular Ca2+buffers in
facilitating and depressing nerve terminals of pyramidal cells
in layer 2/3 of the rat neocortex indicates a target cell-
specific difference in presynaptic calcium dynamics.
J . Physiol. (Lond.)531, 807–826 (2001).
131. Buhl, E. H., Cobb, S. R., Halasy, K. & Somogyi, P. Properties
of unitary IPSPs evoked by anatomically identified basket
cells in the rat hippocampus. Eur. J . Neurosci. 7,
NATURE REVIEWS |NEUROSCIENCE
VOLUME 5 |OCTOBER 2004 |8 0 7
R E V I E W S
132. Pouille, F. & Scanziani, M. Enforcement of temporal fidelity in
pyramidal cells by somatic feed-forward inhibition. Science
293, 1159–1163 (2001).
133. Cobb, S. R., Buhl, E. H., Halasy, K., Paulsen, O. & Somogyi, P.
Synchronization of neuronal activity in hippocampus by
individual GABAergic interneurons. Nature378, 75–78
An excellent paper that revealed the functional impact
of different types of interneuron.
134. Tarczy-Hornoch, K., Martin, K. A., J ack, J . J . & Stratford, K. J .
Synaptic interactions between smooth and spiny neurons in
layer 4 of cat visual cortex in vitro. J . Physiol. (Lond.)508,
135. Segev, I. & Burke, R. in Methods in Neuronal Modeling
(eds Koch, C. & Segev, I.) 93–136 (MIT Press, Cambridge,
136. Segev, I. & London, M. in Dendrites (eds Stuart, G.,
Spruston, N. & Hausser, M.) (Oxford Univ. Press, 1999).
137. Magee, J . C. & J ohnston, D. A synaptically controlled,
associative signal for Hebbian plasticity in hippocampal
neurons. Science275, 209–213 (1997).
138. Larkum, M. E., Kaiser, K. M. & Sakmann, B. Calcium
electrogenesis in distal apical dendrites of layer 5 pyramidal
cells at a critical frequency of back-propagating action
potentials. Proc. Natl Acad. Sci. USA 96, 14600–14604
139. Traub, R. D. Model of synchronized population bursts in
electrically coupled interneurons containing active dendritic
conductances. J . Comput. Neurosci. 2, 283–289 (1995).
140. Tamas, G., Lorincz, A., Simon, A. & Szabadics, J . Identified
sources and targets of slow inhibition in the neocortex.
Science299, 1902–1905 (2003).
141. Thomson, A. M. & Destexhe, A. Dual intracellular recordings
and computational models of slow inhibitory postsynaptic
potentials in rat neocortical and hippocampal slices.
Neuroscience92, 1193–1215 (1999).
142. Benardo, L. S. Separate activation of fast and slow inhibitory
postsynaptic potentials in rat neocortex in vitro. J . Physiol.
(Lond.) 476, 203–215 (1994).
143. Hughes, D. I., Bannister , A. P., Pawelzik, H. & Thomson, A. M.
Double immunofluorescence, peroxidase labelling and
ultrastructural analysis of interneurons following prolonged
electrophysiological recordings in vitro. J . Neurosci. Meth.
101, 107–116 (2000).
144. Kisvarday, Z. F., Beaulieu, C. & Eysel, U. T. Network of
GABAergic large basket cells in cat visual cortex (area 18):
implication for lateral disinhibition. J . Comp. Neurol. 327,
145. Fukuda, T. & Kosaka, T. Gap junctions linking the dendritic
network of GABAergic interneurons in the hippocampus.
J . Neurosci. 20, 1519–1528 (2000).
146. Tamas, G., Buhl, E. H., Lorincz, A. & Somogyi, P. Proximally
targeted GABAergic synapses and gap junctions
synchronize cortical interneurons. Nature Neurosci. 3,
147. Peinado, A., Yuste, R. & Katz, L. C. Extensive dye coupling
between rat neocortical neurons during the period of circuit
formation. Neuron10, 103–114 (1993).
148. Galarreta, M. & Hestrin, S. A network of fast-spiking cells in
the neocortex connected by electrical synapses. Nature
402, 72–75 (1999).
149. Gibson, J . R., Beierlein, M. & Connors, B. W. Two networks
of electrically coupled inhibitory neurons in neocortex.
Nature402, 75–79 (1999).
150. Kisvarday, Z. F. & Eysel, U. T. Functional and structural
topography of horizontal inhibitory connections in cat visual
cortex. Eur. J . Neurosci. 5, 1558–1572 (1993).
151. Borg-Graham, L. J ., Monier, C. & Fregnac, Y. Visual input
evokes transient and strong shunting inhibition in visual
cortical neurons. Nature393, 369–373 (1998).
152. Monier, C., Chavane, F., Baudot, P., Graham, L. J . &
Fregnac, Y. Orientation and direction selectivity of synaptic
inputs in visual cortical neurons: a diversity of combinations
produces spike tuning. Neuron37, 663–680 (2003).
153. Wehr, M. & Zador, A. M. Balanced inhibition underlies tuning
and sharpens spike timing in auditory cortex. Nature426,
154. McBain, C. J . & Fisahn, A. Interneurons unbound. Nature
Rev. Neurosci. 2, 11–23 (2001).
155. Hirsch, J . A., Alonso, J . M., Reid, R. C. & Martinez, L. M.
Synaptic integration in striate cortical simple cells.
J . Neurosci. 18, 9517–9528 (1998).
156. Anderson, J . S., Carandini, M. & Ferster, D. Orientation
tuning of input conductance, excitation, and inhibition in cat
primary visual cortex. J . Neurophysiol. 84, 909–926 (2000).
157. Tan, A. Y., Zhang, L. I., Merzenich, M. M. & Schreiner, C. E.
Tone-evoked excitatory and inhibitory synaptic
conductances of primary auditory cortex neurons.
J . Neurophysiol. 92, 630–643 (2004).
158. Silberberg, G., Wu, C. & Markram, H. Synaptic dynamics
control the timing of neuronal excitation in the activated
neocortical microcircuit. J . Physiol. (Lond.)556, 19–27
The first study to show that subthreshold voltage
cross-correlations correspond to the form of synaptic
159. Galarreta, M. & Hestrin, S. Properties of GABAAreceptors
underlying inhibitory synaptic currents in neocortical
pyramidal neurons. J . Neurosci. 17, 7220–7227 (1997).
160. Varela, J . A., Song, S., Turrigiano, G. G. & Nelson, S. B.
Differential depression at excitatory and inhibitory synapses
in visual cortex. J . Neurosci. 19, 4293–4304 (1999).
161. Bernander, O., Douglas, R. J ., Martin, K. A. & Koch, C.
Synaptic background activity influences spatiotemporal
integration in single pyramidal cells. Proc. Natl Acad. Sci.
USA 88, 11569–11573 (1991).
162. Konig, P., Engel, A. K. & Singer, W. Integrator or coincidence
detector? The role of the cortical neuron revisited. Trends
Neurosci. 19, 130–137 (1996).
163. Mainen, Z. F. & Sejnowski, T. J . Reability of spike timing in
neocortical neurons. Science268, 1503–1506 (1995).
164. Azouz, R. & Gray, C. M. Cellular mechanisms contributing to
response variability of cortical neurons in vivo. J . Neurosci.
19, 2209–2223 (1999).
165. Silberberg, G., Bethge, M., Markram, H., Pawelzik, K. &
Tsodyks, M. Dynamics of population rate codes in
ensembles of neocortical neurons. J . Neurophysiol. 91,
166. Silberberg, G., Gupta, A. & Markram, H. Stereotypy in
neocortical microcircuits. Trends Neurosci. 25, 227–230
167. Toledo-Rodriguez, M., Gupta, A., Wang, Y., Wu, C. &
Markram, H. in The Handbook of Brain Theory and Neural
Networks (ed. Arbib, M.) 719–725 (MIT Press, Boston,
We would like to acknowledge M. Segal, A. Grinvald and
T. McKenna for their long-term support of the work on the micro-
circuit. The studies were supported by a number of grants, including
the Office of Naval Research; Minerva Foundation; Human Frontiers
Science Program; German–Israel Science Foundation; Binational
Science Foundation; Israel Science Foundation; European Union
Fifth Framework; National Alliance for Autism Research; and, more
recently, by the the Swiss Federal Institute for Technology and the
Swiss Science Foundation.
Competing interests statement
The authors declare no competing financial interests.
The following terms in this article are linked online to:
Caα1G | Caα1I| CB | CCK | CR | HCN1 | HCN2 | HCN3 | HCN4 |
Kv1.1 | Kv1.2 | Kv1.4 | Kv1.6 | Kv2.2 | Kv3.1 | Kv3.2 | Kv3.3 |
Kv3.4 | NPY | PV | SK2 | SOM | VIP
The Brain Mind Institute: http://bmi.epfl.ch/
Access to this interactive links box is free online.