Progressin NeurobiologyVol. 41, pp. 281 to 321, 1993 0301-0082/93/$24.

00
Printed in Great Britain. All rights reserved © 1993 Pergamon Press Ltd

PLASTICITY OF THE DENDRITIC SPINE

CATHERINE H . HORNER
Department of Anatomy, Trinity College, Dublin 2, Ireland

(Received 28 May 1992)

CONTENTS
1. The dendritic spine 282
1.1. Introduction 282
1.2. Location and distribution 282
1.3. Shape 285
1.4. Size 286
1.5. Structure 289
1.5.1. Spine apparatus 289
1.5.2. Synapse 289
1.5,3. Cytoplasmic organisation 291
2. Function of the dendritic spine 292
2.1. Introduction 292
2.2. Connectivity 292
2.3. Electronic flow 293
2.4. Calcium regulation 294
2.5. Cytoplasmic proteins 294
2.6. Memory and learning 294
3. Dendritic spine density 295
3.1. Introduction 295
3.2. Methods of assessment 296
4. Plasticity of the dendritic spine 297
4.1. Introduction 297
4.2. Gender effects 298
4.3. Spine morphology 299
4.4. Spine density 299
5. Naturally-occurring plasticity 3OO
5.1. Introduction 3OO
5.2. Temporal variation 3OO
5.2.1. Development 3OO
5.2.2. Aging 301
5.3. Hormonal variation 302
5.3.1. Thyroid 302
5.3.2. Gonadal 302
6. Experimentally-induced plasticity 303
6.1, Introduction 303
6.2, Deprivation 303
6.2.1. Deafferentation 303
6.2.2. Visual deprivation 305
6.2.3. Malnutrition 306
6.2.4. Hypoxia 306
6.3. Stimulation 307
6.3.1. Enriched environment 307
6.3.2. Long-term potentiation 308
7. Plasticity in pathological conditions 310
7.1. Introduction 310
7.2. Disease states 310
7.2.1. Dementia 310
7.2.2. Temporal lobe epilepsy 311
7.2.3. Mental retardation states 311
7,2.4. Sudden infant death syndrome (SIDS) 312
7,2.5. Hypothyroidism 312
7.3. Chemically-induced pathology 312
7.3.1. Ethanol 312
7,3.2. Drugs and toxins 314
8. Conclusions 314
Acknowledgements 315
References 315
Addendum 321

281
282 C. [|. HORNER
I. THE DENDRITIC SPINE
1.1. INTRODUCTION
Dendritic spines are small protrusions or out-
growths of neuronal dendrites which vary in their
density both within the dendritic field of a neuron,
and between neuronal types in different regions of
the central nervous system, Since the discovery by
Golgi of the technique (Golgi, 1873) of impregnating
and staining approximately 10% of cells within the
central nervous system, the structure of neurons
using both light and electron microscopy has been
elucidated in much detail. Golgi may well have seen
examples of dendritic spines on neurons and re-
garded them as artifacts due to the metallic precipi-
tates formed by the Golgi technique, but Cajal
(1891) was the first to describe and publish drawings
of these protrusions admitting their definite existence
as real structures in normal healthy adult cerebral
cortex. At electron microscopic level, Schultz et al.
(1957) studying cerebral cortex, failed to recognise
dendritic spines but their existence has been reconfi-
rmed by both light and electron microscopy (Gray,
1959a,b; Jones and Powell, 1969, Peters and Kaiser-
man-Abramof, 1970). Although identification at the
electron microscopic level can be difficult, internal
features of the spine help to differentiate them from
other neuropil (Section 1.5).
Although dendritic spines vary in shape (Section
1.3) and size (Section 1.4) they are commonly de-
scribed as having a head carrying a synaptic contact
(Section 1.5.2) and a relatively narrow stalk, con-
necting with the dendrite. Sometimes at the site of
synaptic contact of a spine with an axon there are
small protrusions called spinules (Westrum and
Blackstad, 1962; Laatsch and Cowan, 1966) invagi-
hating into the axon. Their internal structure (Sec-
tion 1.5) differs from that of the surrounding
neuropil but the only cytoplasmic organelle regularly
occurring in spines is the spine apparatus which
consists of several flattened or distended cisternae
whose function is, as yet, undetermined although
several suggestions have been made (Section 1.5.1).
The connecting stalk, the spine apparatus, the head
carrying a synaptic contact and the spinule complex
are all structural features characterizing spines.
The function of the dendritic spine is uncertainl as
the many theories (Section 2) suggest. However it
has been well established that the dendritic spine is
an essential part of the postsynaptic membrane of
many types of neurons in the central nervous system
(Cajal, 1911; Gray, 1959a, b; Hamlyn, 1962; Scheibel
and Scheibel, 1968). Gray (1959b) showed that the
spine receives a special class of presynaptic terminal
(Section 1.5.2) while Colonnier's study (1964) indi-
cates that spines are resorbed if the synaptic termi-
nals are lost. The plasticity of dendritic spines to a
diverse group of experimental and naturally-occur-
ring conditions suggest that it is an important struc-
ture in the overall function of the central nervous
system and its response to various phenomena.
1.2. LOCATION AND DISTRIBUTION
Spines are usually found extending from dendrites
(Figs la, b) but are occasionally noted on the
perikaryon, axon hillock or proximal axon of neur-
ons, where they tend to be small and few in number
(Peters and Kaiserman-Abramof, 1970). They
usually arise individually from the dendritic stem,
but a few may arise from a common stalk which
bifurcates into two stalks, each bearing a spine head.
Only rarely does one stalk have two end-bulbs which
arise directly from the common stalk (Peters and
Kaiserman-Abramof, 1970). Each spine head may
contain a spine apparatus and receive an individual
axon terminal (Jones and Powell, 1969).
Spine density is not consistent for all neuronal
types or for all categories of dendrite. Cortical
pyramidal cells are particularly rich in dendritic
spines (Fig. 2a) while non-pyramidal neurons have
relatively few (Cajal, 1911; Scroll, 1956; Globus and
Scheibel, 1967a). There are also regional differences
in Purkinje cell dendritic spines and their parallel
synaptic density (Heinsen and Heinsen, 1983). There
is agreement among researchers that there is a differ-
ential distribution of dendritic spines in relation to
the different parts of the dendritic tree (Westrum
et al., 1964; Marin-Padilla and Stibitz, 1968; Jones
and Powell, 1969; Chan-Palay et al., 1974; Horner
and Arbuthnott, 1991) but disagreement on which
dendrites have the greater density of spines. On
apical dendrites arising at the apex of the perikaryon
there is an initial spine-free zone (Westrum et al.,
1964) although occasionally very small spines are
noted in this area (Globus and Scheibel, 1967a;
Valverde, 1967). This spine-free area is proximal to
the origin of the first secondary dendrites known as
oblique or transverse (Peters and Kaiserman-
Abramof, 1970) and is estimated to be between
20 and 60/~m long (Westrum et al., 1964; Chan-
Palay et al., 1974). Following this spine free area
there is a gradual increase in spine numbers as the
distance from the cell body increases, in cortical
(Westrum et al., 1964), hippocampal (Westrum and
Blackstad, 1962) and Meynert (Chan-Palay et aL,
1974) neurons. This exponential increase has been
noted in mouse visual cortex (Valverde, 1967) and
human motor, auditory and parietal cortices
(Marin-Padilla, 1967). The spine numbers reach
a maximum approximately midway along the
apical dendrite in cortical pyramidal neurons
(Marin-Padilla, 1967; Jones and Powell, 1969;
Peters and Kaiserman-Abramof, 1970) and then
gradually decrease in concentration (Fig. 2b). Golgi
studies of spine distribution in hippocampal pyra-
mids describe a parabolic distribution of spines as a
function of distance from the soma along branches
of apical dendrites (Englisch et al., 1974; Minkwitz,
1976).
The basal dendrites radiating from the base of the
perikaryon also have an initial spine-free zone (Fig. 3)
of 4-15 ~m (Westrum et al., 1964) but like the
secondary and tertiary branches of apical dendrites
they too have many spines. According to several
authors (Jones and Powell, 1969; Westrum et aL,
1964) the apical dendrite has the maximum concen-
tration of dendritic spines while terminal and side
branches of both apical and basal dendrites, which
are of finer diameter, have a lower frequency of
FIc. 1. Dendritic spines protruding from the apical and oblique dendrites of rat hippoeampal (CAI)
pyramidal cells, ap--apical dendrite, ob--oblique dendrite, tt--terminal tuft, arrowheads----dendritic
spines.

283
 !!!~i~

¸¸111!! ~

FIG. 2. (a) the adult human cerebral cortex illustrating spiny pyramidal cells. (b) A montage of an adult
human cortical pyramidal cell illustrating the apical, basal and oblique spiny dendrites, ap--apical
dendrite, b---basal dendrite, o ~ b l i q u e dendrite, arrowheads ~dendritic spines.
284
 PLASTICITY OF THE DENDRITIC SPINE 285

spines. Others disagree, claiming that secondary and
tertiary branches of basal and apical dendrites have
the highest frequency of spines (Peters and Kaiser-
man-Abramof, 1970; Murphy and Magness, 1984;
Homer and Arbuthnott, 1991) noted on different
neuronal types. According to Peters and Kaiserman-
Abramof (1970) the terminal tufts of apical dendrites
have the highest frequency of spines, but Chan-Palay
et al. (1974) claim that the first part of the apical
dendrite has the greatest density of spines although
the basal dendrites and terminal tuft are also highly
spinose.
The differential distribution of the spines through-
out the dendritic field and within a single class of
dendrite is related to the extent of afferent input
the cell receives (Chan-Palay et al., 1974; Marin-
PadiUa and Stibitz, 1968). In the cerebral cortex
this is most obvious where cortical cell lamination
is related to the afferents that terminate there
(Hubel and Wiesel, 1972; Lund, 1973). Attempts to
describe how spines are distributed along the shafts
of neurons by mathematical methods have been
devised (Ruiz-Marcos and Valverde, 1969). A 3-D
mathematical model was used on neurons of mice of
various ages (Ruiz-Marcos and Valverde, 1969) and
this model was validated for the cortical cells of areas
other than vision and in several species including
hamster, cat, monkey and man (Valverde and Ruiz-
Marcos, 1969).
1.3. SHAPE
Dendritic spines have a variety of shapes (Westrum
et al., 1964; Jones and Powell, 1969; Peters and
Kaiserman-Abramof, 1970, Crick, 1982). They have
been classified into three basic shapes (Fig. 3)~(a)
stubby, which are generally short and thick, (b)
mushroom-shaped which have a rather thick stalk
expanding into a bulbous terminal and (c) thin, which
are slender spines which expand into a small oval or
round bulb (Peters and Kaiserman-Abramof, 1970).
Within each of these categories there is some morpho-
logical variation. Jones and Powell (1969) prefer to
describe them in two main types--pedunculated or
sessile. Pedunculated spines are the classical spines
found in the cerebral cortex, with a narrow pedicle
and a cup-like or prism-shaped bulb with a flattened
side receiving the axon terminal at a synaptic contact.
Sessile spines are those with broad pedicles with no
narrowing at the point of contact with the parent
dendrite, but both types vary greatly in shape and
size. In these two basic categories there is variation in
the shape of the spine mainly due to eversion or
invagination of the synaptic surface, or due to differ-
ent points of attachment of the spine pedicle (Jones
and PoweU, 1969). With the exception of the smallest
spines there are three shapes of spine head--rounded,
cup-like and prismatic. It appears that the peduncu-
lated spines with various head shapes classified by
Jones and Powell (1969) correspond to the thin and
mushroom-shaped spines of Peters and Kaiserman-
Abramof's (1970) classification.
The commonest type of spine in cortical pyramidal
neurons is the thin, slender spine (72%). The stubby
spine (19%) is the commonest in the proximal regions
of primary apical or basal dendrites and the mush-
room-shaped spines (9%) have the lowest frequency
(Peters and Kaiserman-Abramof, 1970). Purpura
(1974) found similar types of spines in similar pro-
portions in human motor cortex with thin long spines
noted mainly on basilar and distal apical dendrites
and their branches. Morphological variation of
spines has also been noted in the dentate gyrus
(Laatsch and Cowan, 1966) where the dendritic spine
density of granule cells in the rat hippocampus varies
with the spine shape and location (Desmond and
Levy, 1985). Three forms of spine and three afferent
termination zones have been recorded and statisti-
cally significant differences in spine density were
found among the three spine shapes and variations in

ap
Mushroom

Thin
Stubby
FIG. 3. A camera lucida drawing of the perikaryon and proximal deridritic tree of a rat hippocampal (CAl)
pyramidal neuron demonstrating the spine free zone on the proximal segments of apical and basal
branches and highlighting the commonly described spine shapes, ap---apical dendrite, b--basal dendrite,
oh---oblique dendrite.
286 C, H HORNER
spine density occurred in shape category among the
afferent termination zones.
The functional significance of different shapes or
types of spines has not yet been discerned. Since they
receive afferents which synapse on them it was
thought that the different types of spine might be
responsible for reception of afferents from different
sources (Laatsch and Cowan, 1966). However,
stubby spines are the only category which appear to
be confined to a particular area (proximal end of the
dendrites) (Peters and Kaiserman-Abramof, 1970)
and populations of different spine shapes exist along
a dendrite (Marin-Padilla and Stibitz, 1968; Marin-
Padilla e t al., 1969). In two mammalian species the
distribution of spines along the lengths of apical
dendrites have been resolved into a series of Gaus-
sian curves that correspond to different layers of the
cerebral cortex. However, these results were not
analysed in terms of the morphological varieties of
spines (Marin-Padilla and Stibitz, 1968; Marin-
Padilla et al., 1969). If not related to different affer-
ent inputs the spine shape may be determined by the
size of the axon terminal at the synapse which also
suggests a dependence on the afferent input (Peters
and Kaiserman-Abramof, 1970).
1.4. Size
The majority of Golgi studies suggest that there is
a wide variation in the size of spines both within and
between neurons (Crick, 1982; Jones and Powell,
1969; Marin-Padilla, 1972) although Jacobson (1967)
declared that spines were of relatively uniform di-
mension (approximately 2 p m long) in mammalian
cortex. Crick (1982) claims that spines can be all sizes
on both large and small diameter dendrites, the
tendency being for the smallest dendrites which are
normally secondary and tertiary branches of apical
and basal dendrites, to possess large spines with the
longest pedicles or stalks (Jones and Powell, 1969).
Conversely, spines on the proximal parts of the main
stem dendrites tend to be small. Those on the middle
portions of apical dendrites which appear to be the
main receptive site for extrinsic afferents (Globus and
Scheibel, 1967a, b; Valverde, 1967) are variable but
intermediate between the two extremes (Jones and
Powell, 1969). In Golgi material, the spines on thin
diameter dendrites are of comparable size to the
dendritic diameter and therefore give the impression
of being longer than on thick main stem dendrites
where the spines appear small and inconspicuous
compared with the wide diameter of the dendrites.
Spines on cortical neurons have been estimated to
be up to 2/~m long (Gray, 1959a) with the majority
in the l - 2 p m range (Westrum et al., 1964) while
Valverde (1967) claims that spines in the visual
cortex are up to 3 # m long and Marin-Padilla (1972)
recorded spines of 3 - 5 # m long in human motor
cortex. The spine stalks have been measured as
narrow as 0.1 p m (Gray, 1959a) and up to 1 p m in
diameter (Marin-Padilla, 1972). The diameter of the
spine head has been recorded as between 1 #m
(Valverde, 1967) and 1.5 #m (Marin-Padilla, 1972).
In a study comparing the dimensions of spines found
in the cortical neurons of newborn and adult guinea-
pigs, Schuz (1986) recorded a 21% increase in the
average diameter of the spine head and a 29%
increase in the diameter of the spine stalk in adults
while no difference was noted in spine length and
only a slight correlation between the size of the spine
head and the thickness of the stalk as noted.
Laatsch and Cowan (1966) recorded variations in
the sizes of spines on dentate granule cells in the rat
with the proximal dendrite bearing short spines,
mainly, and the terminal branches having larger ones
which correlates with the observations of Jones
and Powell (1969) on cortical pyramidal spines.
Dendritic spine necks in the dentate gyrus region
are recorded as averaging 0.13pm (Fifkov~i and
Anderson, 1981). Similar values were recorded by
Wilson et al. (1983) using electron microscopic spine
reconstructions.
In the hippocampus Westrum and Blackstad
(1962) found that spines on CA1 pyramidal cells
are usually 1.2-1.4/~m long. The diameter of the
roughly spherical heads is commonly 0.6-1.0/~m; the
elongated heads having small and large diameters in
the range of 0 . 4 / l m - l . 2 # m . The stalks may be as
much as 0.5/~m in diameter, but as a rule are barely
discernible using light microscopy. On electron mi-
croscopy the stalks generally measure 0.3-0.5/~m
long but occasionally up to 1.25/~m or smaller than
0.3#m. The stalks are thinnest in the middle,
measuring about 650-900 ~ in diameter and widen-
ing towards the base and head. The head was
roughly ovoid, frequently 0.4-0.5/~m long with a
transverse diameter of some 0.25/~m the total dis-
tance from the base to top of the spines was most
commonly 0.5-0.6/~m but varied up to 0.8#m
(Westrum and Blackstad, 1962). Similar spine neck
dimensions have also been reported in an electron
microscopic study of the rat CA1 area (Lee et al.,
1980). In a recent study of the hippocampus (Horner
and Arbuthnott, 1991) spines on CA1 pyramidal
cells had an average length of 0.9 #m and a head
diameter of 0.5 # m although the range of dimensions
were wide between individual spines. Spines within
the stratum radiatum appear to be smaller than those
described by Gray (1959a) within the visual cortex
and are more like those noted on Purkinje cell
dendrites which Tavares et al. (1983) estimated to be
1 #m or less under normal circumstances. No sys-
tematic variation was noted for spine dimensions in
relation to either distance from the pyramidal cell
layer or apical versus basilar dendritic regions in
hippocampal neurons but there was considerable
nonuniformity, even between adjacent or nearby
spines (Turner and Schwartzkroin, 1983). They
recorded average dimensions of 0.22pro for neck
width; 0.42 tim for spine neck length and 0.72/~m 2
for spine head surface area for CA3 pyramids.
Obviously, spines of different shapes will have
different dimensions, and therefore spine size will
vary not only between neuronal types but within the
dendritic field. The functional significance of size is
debatable and not really known. Since spines are
classified into categories according to shape, com-
parisons between different studies are difficult unless
one particular type is studied. However, since spines
can alter their shape and size with various stimuli
(Section 6), "the state of activity" of a particular
spine may influence its size.
FIG. 4, An electron micrograph of rat hippocampus (CA1) illustrating a dendritic spine protruding from
its parent dendrite and its synaptic (type 1/asymmetrical) contact with an axon terminal, d--dendrite,
m--mitochondria, s---dendritic spine, t--axon terminal, arrowheads--asymmetrical synapses.

FIG. 5(a). Legend overleaf,

287
FIG. 5. Electron micrographs of rat hippocampus (CA1) illustrating examples of the spine apparatus (arrows). d dcndrit~.
m--mitochondria, s-~lendritic spine, t---axon terminal, arrowheads--synapses (type I or asymmetric),

FIG. 6. An electron micrograph of rat hippocampus (CA 1) illustrating the type 1 (asymmetrical) synapses (arrowheads) with
a dense postsynaptic membrane, d~dendrite, m--mitochondria, t--axon terminal.
288
 PLASTICrrYOFTHEDENDmTICSPXNE
1.5. Saxuca'o~
Dendritic spines are unique structures (Fig. 4)
which are distinguishable from most other neuronal
elements by several features which appear to be
characteristic of them (Gray, 1959a; Seheibel and
Scheibei, 1968). In general, they are devoid of neuro-
tubules and neurofilaments, which are found in den-
drites but frequently contain granular material (Gray,
1959a; Jones and Powell, 1969). However, on oc-
casion, microtubules of the dendritic cytoplasm have
been found in the end-bulb of larger spines (Peters
and Kaiserman-Abramof, 1970; Fifkovfi and Van
Harreveld, 1977). The floccular nature of the granular
material produces a denser spinal cytoplasm than
that of the parent dendrite and sometimes contains a
small cluster of ribosomes which may have a role in
synapse construction and modification (Steward,
1983). Szentagothai et aL (1966) considered the ab-
sence of mitochondria as characteristic of spines but
this is not a constant feature, as large spines may have
mitchondria.
Many spines, particularly large ones, and especially
those located on dendrites as opposed to the
perikaryon, have a cisternal structure or organelle
called the spine apparatus (Section 1.5.1) in the
cytoplasm (Gray, 1959a, b; Jones and Powell, 1969)
and may have up to two or three (Peters and Kaiser-
man-Abramof, 1970). Although unusual, the spine
apparatus has been noted in dendritic shafts (Fifkovfi
and Van Harreveld, 1977). In electron microscopic
preparations, dendritic spines are seen as isolated
profiles of end-bulbs or spine stalks. The bulbs are
usually recognisable due to the presence of a spine
apparatus or synapse site. The spine stalks are more
difficult to identify as they are small rounded profiles
of 0.05~.3 #m, similar to small axons. The absence
of microtubules and the floccular nature of the
cytoplasm with possibly a partial spine apparatus
help to identify them (Peters and Kaiserman-
Abramof, 1970).
1.5.1. Spine apparatus
This organelle, generally found in the spine cyto-
plasm, was first located in the visual cortex of rats
(Gray, 1959a) and later in the hippocampus (Hamlyn,
1962; Westrum and Blackstad, 1962), dorsal horn of
the spinal cord (Gray and Guillery, 1963) and fascia
dentata (Laatsch and Cowan, 1966). It consists of
two or three clear spaces like sacs, vesicles or chan-
nels, approximately 300-500/~ apart (Fig. 5). In
between each pair of opposed membranes of these
"sacs" a dense line, approximately 150-200/~ wide,
occurs (Gray, 1959a; Jones and Poweil, 1969). Gray
(1959a) queried the possibility of the sacs intercom-
municating and Peters and Kaiserman-Abramof
(1970) confirmed that the sacs are not separate enti-
ties but are diverticula and folds of a single complex
cisterna which nearly always communicates with the
cisternae of the smooth endoplasmic reticulum in the
cytoplasm of the parent dendrite. The long axis of the
sacs of the spine apparatus are usually parallel to the
length of the spine stalk.
Gray (1959a) thought that the spine apparatus was
an organelle specific to spines and in fact specific to
289
the neurons of the mammalian forebrain (Gray,
1982). Although usually found within the end-bulb of
dendritic spines, the spine apparatus may extend
beyond the stalk into the dendritic stem (Westrum
and Blackstad, 1962; Jones and Powell, 1969) and
may also occur in the inital segment of axons (Peters
et al., 1968). Although possibly an artifact, its con-
stant association with dendritic spines and type 1
synapses (Jones and Powell, 1969), and the ill-defined
membranous sacs and opaque material in spines
(Colonnier, 1968) confirmed its existence.
The function of the spine apparatus is, as yet,
uncertain but has been considered to be related to the
higher functions of the mammalian forebrain (Ham-
lyn, 1962). Tarrant and Routtenberg (1979) have
shown the presence of filamentous connections and
membranous extensions between the spine apparatus,
the synaptic spinule and the postsynaptic density,
leading them to propose that all these structures are
intimately related to one another. In fact the close
relationship between the spine apparatus and micro-
tubules, the particle-transporting structures, and the
possible continuity between the spine apparatus and
the dendritic agranular reticulum prompted Fifkov~
and Van Harreveld (1977) to propose that these
strucutes provide a path via which newly formed
proteins could reach the spine and so account for the
prolonged volume change in spines following electri-
cal stimulation. The unique structure of the spine
apparatus has led several investigators to propose
that this organdie acts as a repository of both
membrane and protein for the dendritic spine
(Tarrant and Routtenberg, 1979; Rosenbluth, 1962).
It is further proposed that this repository would
provide the necessary material for the mechanism of
synapse division (Carlin and Siekevitz, 1983).
1.5.2. Synapse
The intimate junctions connecting nerve cells--the
synapses--are morphologically specialised contacts
between neuronal processes at which information is
transferred from one neuron to another. The mechan-
ism of transfer is usually by means of chemical
messengers---the neurotransmitters (Shepherd, 1979).
The presynaptic process contains numerous synaptic
vesicles and often mitochondria. The contact region
between pre- and postsynaptic components shows
Iocalised density increases and thickenings (Wyckoff
and Young, 1956; Westrum et al., 1964), with special
adhesive properties (Gray, 1959b). The postsynaptic
density (PSD) is a disk-shaped subcellular organelle
which appears as a thickening of electron-dense
material attached to the postsynaptic membrane of
synaptic junctions (Gray, 1959b) (Fig. 6). Postsyn-
aptic densities may contain one, or more, holes or
perforations giving them a torus or doughnut-shaped
morphology (Cohen and Siekevitz, 1978). The func-
tion of the perforations is unknown but they may
represent an intermediate in the process of synapse
division (Carlin and Siekevitz, 1983). Gray (1959b)
further differentiated synapses into two types--axo-
somatic and axodendritic. Colonnier (1968) also re-
ported differences in synapse types but used different
criteria to that of Gray (1959b).
In type I synapses (Fig. 7) a large percentage of the
290 C.H. HORNER
Spineapparatus
r
Clea sacs
Spine
Microtubules
FIG. 7. A diagrammatic representation of a type ! (asymmetrical) synapse between a prcsynaptic bouton
and a postsynaptic dendritic spine containing a spine apparatus.
length of the opposed membranes is denser and the
postsynaptic thickening is more pronounced than the
presynaptic density. The extraceUular area between
the thickened membranes contains an intermediate
electron-scattering band of material which is usually
nearer the postsynaptic density. Type 2 synapses have
a smaller percentage of thickened opposed mem-
branes with the pre- and postsynaptic thickenings of
similar dimensions and a less obvious intermediate
band. Coionnier (1968) prefers to classify synapses
into asymmetrical or symmetrical types. The asym-
metrical synapse is similar to Gray's (1959b) type 1
with a postsynaptic membrane bordered on the cyto-
plasmic side by a dense, thick opacity corresponding
to De Robertis's subsynaptic web (De Robertis et al.,
1961) and Van der Loos's (1965) subsynaptic or-
ganelle. The symmetrical synapse (Colonnier, 1968)
has no dense, compact, cytoplasmic opacity different
from the membrane and is similar to Gray's (1959b)
type 2 synapse. Both Colonnier (1968) and Jones and
Powell (1969) further define synapse differences based
on the type of vesicle population. Spheroidal or
slightly oval vesicles of relatively homogenous size are
the commonest in the cerebral cortex and are strongly
correlated to asymmetrical synapses. In the sym-
metrical type of synapse there is a mixture of vesicles
which can be flattened or pleomorphic or small and
round. Despite the imperfect analogy between the
two types of synapses, it is generally agreed that type
I or asymmetrical synapses are found mainly on
dendritic spines (Gray, 1959b; Colonnier, 1968; Jones
and Powell, 1969). Type 2 or symmetrical synapses
are usually located on the neuronal soma or dendritic
trunk (Lund and Westrum, 1966). Although there has
been criticism of the classification of synapses into
two clearly different types (Van der Lees, 1965),
Colonnier (1968) has confirmed that in the main,
synalr'~es fit easily into one of the two types of
synaptic membrane junction.
Type 1 synapses are thought to be excitatory while
type 2 are inhibitory (Raviola and Raviola, 1967)
and, similarly, spheroidal vesicles are related to
' ~ "
Vesicles
t~ I Mitochondrla
Pre-synaptic membrane
excitatory synapses, while flattened vesicles are
associated with inhibitory synapses (Larremendi
et al., 1967; Uchizono, 1966). Hence dendritic spines
have excitatory contacts (Colonnier, 1968; Gray,
1959b, Eccles, 1964).
In cerebral and cerebeilar cortex, only a limited
area of the dendritic spine plasma membrane
(7-10%) is occupied by the synaptic active zone
(Spacek and Hartman, 1983). In their study they
noted linear dependencies of dendritic spine surface
area on dendritic spine volume, and synaptic active
zone surface area on dendritic spine surface area
which implies that the synaptic active zone to
dendritic spine surface ratio may be functionally
important.
Usually each spine has one synapse with one axon
terminal from extrinsic afferents, the asymmetrical
synapse being on the end-bulb (Jones and Powell,
1969; Peters and Kaiserman-Abramof, 1970; Crick,
1982). However, a spine may synapse with two or
more axon terminals, the second synapse being on-the
stalk portion of the spine (Peters and Kaiserman-
Abramof, 1970; Colonnier, 1968; Jones and Powell,
1969). This second synapse site occurs in 10-20% of
spines and can be either type I or II and may even be
located on the parent dendrite and contain a second
spine apparatus (Jones and Powell, 1969). The size of
the synaptic junction between the spine and the axon
terminal is variable depending on the size of the
end-bulb. Thin and stubby spines have small end-
bulbs with simple junctions of one or two postsyn-
aptic densities. Mushroom-shaped spines have
extensive junctions with three or four postsynaptic
areas, the axon terminals often bulging into the
end-bulb (Peters and Kaiserman-Abramof, 1970).
The edge of each synapse is surrounded by an
astrocyte or axon which insulates the synapses (Peters
and Palay, 1965).
Within the dendritic spine bulb, a structure termed
the spinute-complex, consisting of a small round or
oval profile surrounded by a larger profile of the same
shape, has been noted (Westrum and Biackstad,
 PLASTICITYOF THE DENDRITIC SPINE
1962). The outer membrane is an invagination of the
cell membrane of the terminal and the inner profile a
projection (spinule) from the head or stalk of a spine.
Spinule complexes vary in size, but invaginations are
commonly 1,000-2500 A long, 600-1000 A wide and
the spinules are correspondingly smaller (750-1500 A
long, 250-1000A wide). They do not contain the
dark substance seen in the subsynaptic cell membrane
of synapses and do not appear to arise from that area
of the dendritic spine but often arise from a point
close to such a site or even can be surrounded by the
dark subsynaptic area. The "spinule complexes" oc-
curring only at axo-dendritic junctions may be related
to impulse transmission and may not be permanent
structures (Westrum and Blackstad, 1962).
The synaptic junction varies in size and may have
more than one PSD visible, depending on the plane
of section--sections at right angles to the synaptic
junction showing multiple densities. The apparently
separate postsynaptic densities are portions of a
single dense plate with small perforations (Peters and
Kaiserman-Abramof, 1970). Geinisman et al. (1987)
recorded that a proportion of axospinous synapses
exhibit a perforated postsynaptic density and this is
a general phenomenon shared by different types of
synapses. Changes in the proportion of synapses
containing postsynaptic densities with perforations
during periods of increased synapse formation have
led to a hypothesis suggesting the possible division of
pre-existing synapses (Carlin and Siekevitz, 1983).
They found that various types of stimulation result in
an increased synaptic junction area where a perfor-
ation forms in the enlarging synaptic junction. A
synaptic spinule may then appose the perforation in
the postsynaptic density which increases in size until
the synaptic junction splits into two separate junc-
tions within the same synaptic terminal. The dendritic
spine then divides, each half containing a synaptic
junction. This theory is supported by the fact that
increases in synapse number correlate with an in-
crease in the proportion of synapses containing post-
synaptic densities with perforations (Greenough
et al., 1978; Hattan and Ellisman, 1982); probable
intermediates for synapse division have been seen
with horseshoe-shaped or dumbell-shaped junctions
appearing as two separate junctions in the same
terminal (Vrensen and Nunes-Cardozo, 1981) and
during lesion-induced increases in synapse number
there is a large increase in the number of presynaptic
terminals making separate synapses on two or more
separate processes of the same dendrite (Matthews
et al., 1976b; Field et al., 1980). Furthermore, in-
creased numbers of dendritic spines resulting from
rearing in a complex compared to a deprived environ-
ment (Volkmar and Greenough, 1972; Globus et al.,
1973; Valverde, 1971) are found to have an increased
percentage of PSDs with perforations. Therefore the
perforation in the PSD may be related to the for-
mation of new synapses located on the new spines
(Greenough et al., 1978). It is difficult to explain how
the just-divided synapse can transform into two
synapses on two dendritic spines, but y-shaped den-
dritic spines have been observed (Wilson et al., 1983),
and one possibility is that the dendritic spine could
split in two. Actin (Section 1.5.3) could provide the
contractile force for this type of process because actin
291
has been shown to be present in very high concen-
trations in the dendritic spine (Fifkovfi and Delay,
1982). This model of synapse division is attractive
beacuse it provides a function for three speeialis-
ations (PSD perforation, synaptic spinule and spine
apparatus) for which no definite functions have yet
been shown to exist, except for the calcium sequester-
ing ability of the spine apparatus (Burgoyne et al.,
1983).
1.5.3. Cytoplasmic organisation
The cytoplasmic organization of dendritic spines
was studied in Purkinje cells of the cerebellum and
three sets of filamentous structures were found associ-
ated with synaptic junctions (Landis and Reese,
1983). The dendritic spines were filled with a mesh-
work of filaments (5-7 nm) which contacted the spine
membrane except at the synaptic junction, and ex-
tended through the spine neck to the parent dendrite.
The PSD subjacent to the postsynaptic membrane
consisted of a web of filaments (4-6 nm), while a
group of larger microfilaments (8-10 nm), possibly
made of actin, was associated with the postsynaptic
web and extended into the rest of the spine. The
organisation of these filament structures may account
for the shape of the spine. Furthermore fine filaments
and irregular globular adherent proteins have been
located in the substructure of the PSD (Landis et al.,
1987). The actin-like microfilaments and spectrin-like
filaments are juxtaposed to the postsynaptic density
but apparently are not continuous with the constitu-
ent filaments of the density, which suggests that the
PSD is composed of fine filamentous proteins that
insert on the postsynaptic membrane. The filaments
may form a supporting framework for globular pro-
teins which are found in the area of the spine with the
highest concentration of ionised calcium entering
with the synaptic current and the greatest extent of
postsynaptic depolarisation (Landis et al., 1987).
In further studies, proteins, and in particular the
contractile protein-actin, have been localised in hip-
pocampal (Drenckhahn et al., 1984; Moshkov et al.,
1986) and cerebral cortex neurons (Cohen et al.,
1985). Actin expresses a preferential affinity for the
main synaptic areas, approximately the size of synap-
tic structures i.e. dendritic spines and presynaptic
terminals, suggesting a role for actin in synaptic
function (Drenckhahn et al., 1984). In fact Cohen
et al. (1985) found that actin in the PSD was continu-
ous with the subsynaptic web filaments which in turn
was in continuity with the spine apparatus and or the
synaptic membranes adjacent to the PSD. Therefore,
events at the synaptic zone may communicate from
the postsynaptic membrane via the PSD to the cyto-
skeletal network--the subsynaptic web. Conse-
quently actin present in dendritic spines may play a
role in changing the spine shape and synaptic curva-
ture (Cohen et al., 1985; Moshkov et al., 1986).
Actin filaments have also been located throughout
dendrites in the developmental period and the highest
concentrations were found in regions of spine devel-
opment (Markham and Fifkov/l, 1986), the actin
filaments within them increasing in number as the
developing spines became more complex. Therefore
their hypothesis, that actin plays a role in the
292 C.H. HORNEa
protrusion of spines from the dendrite, seems plaus-
ible. Kimura et al. (1987) also studied the postnatal
development of the membrane associated cytosketetal
protein----calspectrin--in terms of expression and
localisation in visual cortex neurons in rats. Calspec-
trin was detectable at birth at the plasma membrane
and in the cytoplasm of increasing numbers of differ-
entiating neurons until spines were visualised. Adult
levels of immunoreactive calspectrin were very low
and spines could not be recognised. Therefore, cal-
spectrin is most abundantly expressed in the growing
parts of dendrites and spines which suggests that
calspectrin may play a role in synaptic plasticity in
development. Fifkov~i and Morales (1989) agree that
the calcium-regulated contractile and cytoskeletal
proteins in dendritic spines may control synaptic
plasticity.
2. FUNCTION OF THE DENDRITIC SPINE
2.1. INTRODUCTION
The function of dendritic spines is the subject of
much debate. Since the first recorded observation of
dendritic spines by Cajal (1891) it has been repeatedly
confirmed that they are sites of synaptic contact
(Gray, 1959a,b; Colonnier, 1968; Boycott, 1982). In
191l, Cajal suggested that their principal function
was to increase the receptive area of the dendrites and
hence permit a greater number of synaptic contacts.
However, although they provide a high degree of
connectivity (Section 7.2) this alone does not explain
why spines are situated on dendrites and not on
axons, nor why the spine cytoplasm differs from that
of dendrites and contains a spine apparatus which
Gray (1982) maintains to be specific to the spines of
the mammalian forebrain. A structure similar to the
spine apparatus has been located in the initial axon
segments of pyramidal neurons. In both locations the
cisternae of the apparatus are confluent with the
smooth endoplasmic reticulum and therefore suggests
that it is concerned with synaptic activity and seques-
tering ions (Peters et al., 1968). The ability of the
spine apparatus to sequester calcium ions (Burgoyne
et al., 1983) and the proposed role of dendritic spines
in regulation of calcium ion dynamics contrasts with
traditional approaches to spine function that have
stressed electronic properties (Zador et aL, 1990).
As spines are thought to have excitatory contacts
only (Eccles, 1964; Gray, 1959b) while synapses on
dendritic trunks or soma may be either excitatory or
inhibitory, Gray (1982) suggests a possible role in the
selection of axons during ontogenesis. In ontogenesis
the inhibitory axons to other parts of the cell must
pass close to the spine which may produce a signal
recognised only by excitatory axons. However, if
spines have excitatory asymmetrical synapses with
extrinsic connections, the secondary synapse with a
symmetrical inhibitory terminal en passant with in-
trinsic afferents, sometimes present, possibly provides
an excitatory-inhibitory interaction within a single
spine (Jones and Powell, 1969).
There must be a relationship between neuronal
structure and function. The geometry and electrical
properties of neurons determine how synaptic inputs
and endogenously-generated currents are integrated
and transformed into signals transmitted to other
cells (Miller and Jacobs, 1984). The electrical proper-
ties of spines are important since it has been shown
that electrical stimulation of neurons produces both
short- and long-term volume changes in the spine
(Fifkovfi and Van Harrefeld, 1977). The contractile
proteins located at higher concentrations in spines
(Cohen et al., 1985; Moshov et al., 1986) than in other
neuronal areas may enable alterations of spine shape
to occur (Crick, 1982). All these features contribute
to the suggestion that the dendritic spine is related to
higher functions in mammals and in particular, learn-
ing and memory (Hamlyn, 1962).
2.2. CONNECTIVITY
Because dendritic spines bear the majority of
synapses (Colonnier, 1968; Marin-Padflla, 1968;
Crick, 1982) it is accepted that the spines are import-
ant sites of synaptic contact (Boycott, 1982; Gray,
1982; Swindale, 1981). In fact, Swindale (1981) postu-
lated that dendritic spines are primarily morphologi-
cal devices for connecting axons and dendrites.
Increased density of spines is seen where more con-
nections are required between pre- and postsynaptic
cells. Because the synaptic density in the cerebral
cortex is close to its upper limit, almost all synapses
formed by spines must be with axons en passant
(Braitenberg, 1978). Similar conditions of connec-
tivity occur in the cerebellum and hippocampus
(Shepherd, 1979). Peters and Kaiserman-Abramof
(1970) agree that at least for apical dendrites the axon
terminals running parallel to the dendrite provide
boutons en passant. This explanation may be too
simple in that it does not explain why spines are
located on dendritic trunks as opposed to the soma,
nor why other neural connections are made at
synapses without the presence of spines, If spines are
removed, there is a reduced number of connections
and specificity of connections can only be maintained
at the expense of a drop in synapse density, or vice
versa (Swindale, 1981). A solution to the problem of
reduced spines is possible if axons and dendrites
could be rerouted. This would be simple for axons
which could grow a short distance along the track of
the previously existing spine. However, intervening
axons and dendrites would exist so that reconnection
would require substantial changes in organisation of
the neuropile (Swindale, 1981). If Swindale's (1981)
hypothesis is correct and spines do not serve any
additional function they should be absent in some
circumstances such as when a high convergence of'
connections occurs as demonstrated between basket
and Purkinje cells where synapses are made directly
with the cell body and proximal shaft (Shepherd,
1979); even though they have the ability to form
spines. In addition, spinal motor neurons lack spines
but receive inputs from small numbers of presynaptic
cells, each of which may make many contacts with a
single motor neuron. Therefore, spines are absent in
situations where they are not necessary--they are
mainly connectors between nonadjacent axons and
dendrites (Swindale, 1981). Gray (1982) agrees with
Swindale (1981) that they provide sites of connection,
however, there are extensive areas of dendrites
 PLASTICITY OF THE DENDRITIC SPINE
between spines which are not covered with axon
terminals and synapses (Jones and Powell, 1969;
Colonnier, 1968; Peters and Kaiserman-Abramof,
1970; Gray, 1982). Gray (1982) points out that these
areas are important as sites of close apposition of
astrocytes. He suggests that a further function of the
spines could be to "lift" presynaptic contacts so that
areas of dendrite are free to come into close relation-
ship with, and interact with, neuroglia. Peters and
Kaiserman-Abramof (1970) also suggest that the
spines lift the synapses, and thus enable the dendrite
to reach out to the axon rather than changing course
and zig-zagging to reach the synapse site. This is a
very efficient means for one axon to form several
synapses with different neurons such as occurs in the
cerebral cortex, where the axons run parallel to the
apical dendrites (Marin-Padilla, 1968).
2.3. ELECTRONIC FLOW
Chang (1952) coined the phrase "synaptic weight"
and stated that the electrical resistance of the neck of
the spine could reduce the "weight" of the synapse,
meaning the electrical effect which a presynaptic
activation of the synapse has on the impulse initiation
site of the cell. This is not a straightforward attentu-
ation effect but is due to the fact that synaptic
activation involves a transient change of conductance
at the synapse (Rinzel and Rall, 1974) and the extra
resistance of the spine neck slows the resultant flow
of ions during synaptic activation. Subsequently Rall
(1978) pointed out that the impedance of the spine
tends to match that part of the dendrite to which it
is attached, since spines with long narrow necks are
found more often on the distal parts of dendrites.
More sophisticated calculations by Koch and Poggio
(1983) have confirmed the theory of Rail and Rinzel
(Rail and Rinzel, 1973; Rinzel and RaU, 1974) that
plausible values for a change of shape could alter the
weight of the synapse by a factor of two or three.
Recent application of cable (Turner, 1984a;
Wilson, 1984) and core-conductor theory to the
anatomical study of spines has led to new ideas on
function (toss and Perkel, 1985). One theory suggests
that if spines are treated as tiny dendrites with much
higher resistances than the larger parents, then the
high spine-stem resistance results in relative electrical
isolation of the spine head which produces a large
local depolarisation in the spine head. However, in
studying the electrical properties of dendritic spines in
neurons with arbitrary geometry, Kawato and
Tsukahara (1983) treated all dendritic branches, in-
cluding spines, as cyclinders of uniform passive mem-
branes and showed that the postsynaptic potential
due to the synapse on the spine is representative of
two functions. The first is the postsynaptic potential
caused by the same synapse on the branching point
where the spine stalk is attached to the main dendritic
trunk. The second is determined by the morphologi-
cal and electrical properties of the spine and it
represents the attenuation effect of the spine. On the
assumption that the diameter of the spine stalk is
sufficiently small compared to its parent dendrite, an
approximation of the second function was obtained.
Kawato and Tsukahara (1983) concluded that
synapses on dendritic spines are not effectively
293
isolated from other synapses since morphological
changes to the spine did not produce an effective
change of postsynaptic potential.
Anatomical evidence indicates that, particularly
for discrete dendritic inputs, a detailed segmental
model may be more appropriate than the equivalent
cyclinder model. Turner (1984a) produced such a
cable model of neuronal structure, to predict the
effects of discrete transient inputs on hippocampal
neurons (CA1, CA3 pyramids and dentate granule
cells). The dendritic shaft and spine input impedance
was computed for sites on hippocampal neurons and
the spine input impedance for a typical spine ap-
pended onto a dendritic shaft averaged less than 2%
higher than the dendritic input impedance values for
the adjacent dendritic shaft (Turner, 1984b). The
spine synaptic inputs, simulated by a brief conduc-
tance transient, demonstrated that the average spine
transient was attentuated by less than 2% in conduc-
tion across the spine neck. The same input transient
applied to the proximal and distal sites on CAI
pyramidal cells demonstrated a clear difference be-
tween proximal and distal inputs at the soma as the
predicted result. Therefore, the main determinant of
passive propagation of transient electrical signals in
these neurons appeared to be the dendritic branching
rather than the signal attentuation through the spine
neck. Impulse propagation toward the cell body
along the dendrites is slow and decremental and the
probability that a neuron will be discharged is mini-
mal when synapses are located in the most distal parts
of the dendrites (Eccles, 1964).
Another theory proposes that if the spine head
input resistances are higher than those of the parent
dendrites, spines have the potential for modulating
many biochemical and biophysical processes that
might regulate synaptic efficacy (Coss and Perkei,
1985). In a further cable model of the linear proper-
ties of dendritic spines, Wilson (1984) simulated the
effect of a synaptic conductance change by using
analytical solutions for the voltage generated in the
spine by a current impulse at the spine head. Synaptic
current produced by the conductance change was
used as an input for evaluation of the postsynaptic
potential and current injected into the dendrite at the
base of the spine. Wilson (1984) found that the
primary effect of the dendritic spine was to attentuate
synaptic current which was produced by high input
impedance at the axospinous synapses, which re-
suited in giant spike-like excitatory postsynaptic po-
tentials that approached reversal potential of
synapses and therefore lowered the potential gradient
draining the synaptic current. Although virtually all
synaptic current was transferred to the dendrite, it
produced much smaller excitatory potentials there,
due to low dendritic input impedance. Very small
conductance changes produced near-maximal synap-
tic currents in dendritic spines. The current-attentuat-
ing effect of the spine was accentuated with brief
synaptic transients and decreased with prolonged
synaptic conductance changes. The size and shape of
the spine head, and the dendritic diameter had little
or no effect on the current attentuation for spines in
the "normal" size range. The diameter and length of
the spine stalk and the size and location of the
spine apparatus were the key morphological factors
294 ~i 1"1.HORNER
determining the synaptic currents generated by axo-
spinous synapses. Naturally occurring size and
shape differences among dendritic spines produced
large differences in synaptic potency when compared
in a model spiny neuron. The differences were com-
parable to those produced by differences in synaptic-
location on the same neuron.
Horwitz (1984) proposed a postsynaptic mechan-
ism for transforming electrical activity at the
synapse into structural modifications near the
synapse while studying the electrophoretic migration
of charged metabolites. The potential difference be-
tween a postsynaptic potential generated at the
synapse and the attenuated result gives rise to an
intraneuronal electric field, and allows the electro-
phoretic migration of charged metabolites, both
along the inner surface of the membrane and in the
cytoplasm. Applied to dendritic spines, theoretical
calculations based on this mechanism show that
synaptic activity at a spine head can establish large
electric fields along the spine, whereas synaptic ac-
tivity at the base of the spine generates very small
electric fields in the spine---the thinner the spine, the
stronger the field.
Empirical studies have noted that spine heads
increase rapidly in size after afferents are stimulated
electrically when animals have engaged in a bout of
important behavioural activity (Coss and Perkel,
1985). The spine head enlargement dilates a portion
of the spine stem without much alteration in the
overall spine length. Further theoretical study
shows that spine-stem shortening decreases spine
head input resistance relative to the branch input
resistance. Decreased input resistance can enhance
the transfer of electrical charge from the spine
head to the parent dendrite, especially when
synaptic conductance is large relative to the spine
head input conductance. Spine-stem shortening can
also decrease the peak transient membrane potential
in the spine head and this could delimit calcium
ion influx into the spine head via voltage-dependent
calcium channels. Modulation of calcium influx
by spine-stem shortening has the potential for
regulating calcium sensitive enzymatic activity in
the spine head that could affect phosphorylation
of cytoskeletal proteins maintaining spine shape
and phosphorylation of proteins in the post-
synaptic density. Kawato and Tsukahara (1984)
investigated the transient electrical behaviour of
spines with bulbous terminals in neurons with
arbitrary dendritic geometries. They noted that
when the synaptic resistance is comparable to the
spine-stem impedance a morphologic change in
the spine can induce changes in synaptic current
and the postsynaptic potential due to the so-called
nonlinear effect to the synapse. After corticorubral
stimulation of rubrospinal neurons Kawato et al.
(1984) noted that as the spine shortens, the
amplitude of the excitatory postsynaptic potential
becomes considerably larger, but the time-to-
peak value does not change markedly. If the
unitary excitatory postsynaptic potential in
a rubrospinal cell is produced by activation of
several terminals, a morphological change of the
spine has a smaller effect on the excitatory postsyn-
aptic potentials.
2.4. CALCIUMREGULATION
Until recently, theoretical investigations of spine
function focused on the large electrical resistance
provided by the narrow constriction of the spine
neck. However, the narrow constriction is also
thought to provide a large diffusional resistance.
When calcium currents were activated in spines, the
peak spine head calcium ion concentration was
greater in "long, thin" spines than in mushroom-
shaped or stubby spines. The same currents, activated
on dendrites, produced even smaller local calcium
concentration changes (Holmes, 1990).
Zador et al. (1990) devised a biophysical model of
electrical and calcium ion dynamics which accounts
for much of the phenomenology of the induction of
LTP at a Hebbian synapse in the hippocampus
(CA1). Computer simulations suggested four import-
ant functions of spines in this calcium-dependent
synaptic modification.
(1) Compartmentalising transient changes
in calcium concentration to just those
synapses that satisfy the conjunctive re-
quirement for synaptic modification.
(2) Isolating the spine head from changes in
calcium concentration at the dendritic shaft.
(3) Amplifying the concentration changes
at these synapses.
(4) Increasing the voltage dependence of the
processes underlying LTP induction.
Previously Gamble and Koch (1987) noted that
short high-frequency bursts of presynaptic activity
were more effective in raising levels of calcium and
especially the calcium-calmodulin complex, than sus-
tained low activity. Given the importance of calcium
for long-term potentiation (LTP), the ability of spines
to concentrate calcium ions may play a key role in
processes leading to learning and memory storage.
2.5. CYTOPLASMICPROTEINS
If the cytoplasm of spines contains a reasonable
concentration of proteins such as actin or myosin
(Cohen et al., 1985; Moshov et al., 1986), the time
needed for a muscular gel of this type to contract
and the time for the ions or small organic molecules
to diffuse over the small distances involved are short
enough for the mechanism proposed by Crick (1982)
to work. He suggests that if actin is present in high
concentrations in spines but absent, or rather low, in
the adjacent dendritic shaft, spines may be capable
of twitching rapidly during normal neural activity.
Therefore, contractile proteins associated with the
spine cytoplasm, could allow the spine to change its
shape rapidly during neuronal activity. Therefore,
heavily spinous neurons such as pyramidal or stel-
late cells are not static objects, since many of its
spines are changing their shapes in response to
synaptic inputs.
2.6. MEMORYAND LEARNING
The dendritic spine has been implicated in the
mechanism of learning and memory from earlier
times. Cajal (1911) and Hebb (1949) speculated that
long-term memory storage might involve changes in
 PLASTICITY OF THE DENDRITIC SPINE
synaptic size and number. Kandel and Spence (1968)
postulated that permanent changes may occur in
synapses as a result of functional activity and that
these may play some part in learning and memory.
Others have suggested the opposite possibility--that
learning involves selective elimination of neurons and
or synaptic connections (Cajal, 1929; Jacobson, 1970;
Dawkins, 1971). Eccles (1965) suggests that growth of
those synapses already present are involved in learn-
ing and memory while Sperry (1965) doubts whether
morphological changes have any role.
Rail and Rinzel (1973) suggested that memory
might be stored in the shape of the dendritic spines.
A difficulty with this idea is that spines do not appear
to be rigid in structure but vary in shape and size as
indicated by Fifkov/t and Van Harrefeld (1977), when
they demonstrated long- and short-term volume in-
creases in spines following stimulation. Long-term
memory in the human may last for decades and it
might be expected that it resides in a more rigid
structure, such as the synaptic junction itself, so that
the old could be replaced by new molecules without
disturbing the overall structure (Crick, 1982). Long-
term memory may be stored in a distributed and
redundant way, as the "weights" of many synapses
(Section 2.3). Therefore a new memory could be
recorded as a small alteration to the "weights" of
many synapses (Crick, 1982). Various rules have been
proposed for the manner in which synaptic weights
are altered, a popular one being that a synapse is
strengthened (i.e. the synaptic "weight" is increased)
if the presynaptic activation of the synapse is approxi-
mately coincident with the firing of the cell postsyn-
aptic to the synapse (Hebb, 1949). The various rules
share in common the fact that the weight of each
synapse is either increased or decreased depending on
the correlation between the presynaptic activation
and the postsynaptic activity. This type of rule would
provide a great advantage in the efficient functioning
of higher nervous systems (Crick, 1982). Short-term
memory is thought to depend in some way on
reverberating electrical circuits or resonance but it is
usually assumed that such memory is expressed
mainly by the firing pattern of complex sets of
neurons and involves little or no change in synaptic
weights, which are only altered appreciably when
long-term memory is laid down (Crick, 1982). If this
chain of ideas has any validity then it is obvious that
the neck of the dendritic spine is ideally suited to
perform this function (Section 2.3). By altering its
shape it can contribute a suitable term to the synaptic
weight. The contraction of an elongated gel is likely
to make it shorter and fatter and thus decrease the
electrical resistance of the neck (Section 2.5). The
position of the neck is such that it could, by one
mechanism or another, be influenced both by its own
synapse and by any very recent activity of neighbour-
ing synapses (Crick, 1982). The electrical properties,
the movement of charged metabolites, the cyto-
plasmic contractile proteins and the plasticity of
shape and size of the dendritic spine all appear to
interplay to produce the overall function of the
dendritic spine.
Recent experiments support the view that dendritic
spines are involved in memory formation processes
(Patel and Stewart, 1988) and hence since they are the
295
main synapse bearers, synaptic connections are most
likely involved. Following one-trial passive avoidance
training, the spine density of neurons in the chick
intermediate hyperstriatum ventrale was increased by
89-113% in the left cerebral hemisphere and by
37-69% in the right. Although minor changes oc-
curred in spine dimensions there was no overall effect.
The spine shape and density changes occurred rapidly
(25 hr) following training. The hemispheric assymme-
try in the control chicks (47% higher spine density on
the right side) was abolished after training. In a
further study, Patel et al. (1988) gave a subconvulsive
transcranial electroshock to the chicks 5 rain after
training, which rendered half of them amnesic. In
those chicks that remembered the response, the spine
density was significantly increased 24-30 hr after
training in the left intermediate medial hyperstriatum
ventrale when compared with those rendered am-
nesic. This argues for a specific role for dendritic
spines in memory formation in chicks (Patel et al.,
1988).
Synapse division provides a mechanism for func-
tional reorganisation and may well be the method by
which the increase in spine density occurs in memory
formation processes. Repetitive stimulation could
induce a greater number of synapses per neural
pathway via synapse division, thereby enhancing the
ability of the first neuron to generate an action
potential in the second neuron. Carlin and Siekevitz
(1983) describe a possible division of preexisting
synapses during periods of increased synapse for-
mation. This model suggests that various types of
stimulation result in a sequence of events leading to
division of the dendritic spines, with each subdivision
having the ability to function as a synapse. Physio-
logical responses in which synapse division may
possibily play a role include hormone-induced neur-
onal changes, reinnervation of dendrites after lesions
and learning and memory.
3. DENDRITIC SPINE DENSITY
3.1. INTRODUCTION
The functional significance of dendritic spines and
their morphological sensitivity to a wide spectrum of
experimental and pathological states has led to many
studies in which dendritic spine density estimates
have been performed. Spine density is equal to the
number of dendritic spines per micrometre of den-
drite. Some studies claim that primary dendrites have
lower spine density than secondary or tertiary
branches (Peters and Kaiserman-Abramof, 1970;
Murphy and Magness, 1984; Homer and Arbuthnott,
1991) while others maintain that large diameter pri-
mary dendrites have the greater spine density
(Westrum et al., 1964; Jones and Powell, 1969).
Spine density increases with development from
birth until a peak is reached and then falls to adult
levels (Murphy and Magness, 1984) but the rate and
pattern of increase are not uniform for all dendritic
branches and spine density changes differ for specific
cell types. Secondary and tertiary branches differ
from primary branches in reaching their peak later
and in stabilising at an adult level greater than that
296 C.H. HORNER
for primary dendrites (Murphy and Magness, 1984).
Therefore, spine density is a function of age and
although it cannot be assumed that spines in imma-
ture neurons necessarily make functional synaptic
contacts (Sotelo, 1975), spine development is a major
feature of postnatal maturation of cortical neurons.
Such a time-course correlates well with indices of
functional maturation (Boothe et al., 1979; Juraska
and Fifkov~i, 1979). The characteristic alteration in
spine density with age may correlate with other
factors determining the plasticity of an individual
neuron's functional determination. In some cortical
areas a plateau follows while in others a significant
decrease occurs from peak to adult levels.
The assumption that dendritic spines are postsyn-
aptic structures, which, in heavily spined neurons
such as the cortical pyramidal cell, mediate the
majority of the neuron's synaptic input (Colonnier,
1968) has led to many studies in which dendritic spine
counts are determined as a feature of neuronal mor-
phology and cell type differentiation (Feldman and
Peters, 1979). The change in dendritic spine density to
experimental (Valverde, 1967; Globus and Scheibel,
1967b,c), pathological (Marin-Padilla, 1974; Pur-
pura, 1974) and naturally-occurring conditions (Feld-
man and Dowd, 1975) has also resulted in dendritic
spine quantification being used as a good indicator of
neuronal plasticity.
3.2. METHODSOF ASSESSMENT
Dendritic spine density can be assessed by several
methods. Until recently, spine density was estimated
by counting the number of visible spines along a
length of dendrite which was also estimated using an
ocular micrometer (Eqn 1). Many studies have
recorded spine densities of apical, basal and oblique
or transverse and terminal tuft dendrites in a variety
of cell types. This type of estimate is a useful indicator
of spine density within the same dendritic type of the
same cell type from the same species. In its favour, is
the fact that it is a simple measurement performed by
light microscopy involving less time than more soph-
isticated methods. However it has several drawbacks
when comparing densities across dendritic or cell
types or species (Horner and Arbuthnott, 1991). This
simple method of estimation totally under-represents
the true density of spines on any dendrite (Chan-
Palay et al., 1974; McConnell and Berry, 1978;
Feldman and Dowd, 1975) in that, due to the cyclin-
drical shape of dendrites most spines are made invis-
ible by the dark impregnation of the Golgi method
used for these quantitative studies. Since dendrites
vary in diameter, thicker ones are underestimated to
a greater extent than those of smaller diameter,
although this to some extent depends on the distri-
bution of spines being similar throughout the
dendritic field. Therefore, comparisons between den-
drites, or even between segments of the same dendrite
are not always valid. The tortuosity of the dendrite
over which the spines are counted also makes it
difficult to acheive an accurate measurement of den-
dritic length. Other than slow serial electron mi-
crosopic sectioning and rebuilding of spines, no
efficient and relatively accurate assessment of spine
density was available until several researchers (Chan-
Palay et al., 1974; Bradley and Berry, 1976; Feldman
and Peters, 1979) developed geometrical equations
(Eqns 2-4) in an attempt to overcome the underesti-
mation with visible spine counts.
Equation I
Dendritic spine density = n/Dl
where n = number of visible spines, DI = dendritic
length over which spines are counted
Equation 2 (Feldman and Peters, 1979)
N = nn[(Dr + SI): - (Dr + Sd)2]/
[O/90r~(Dr + SI) z] - 2[(Dr + Sl)SinO(Dr + Sd)]
where N = an estimate of the total number of spines,
n = number of visible spines, Dr = dendritic radius,
SI = spine length, Sd = diameter of spine head,
Cos0 = Dr + Sd/Dr + SI, DI = dendritic length over
which spines are counted. Therefore: Dendritic spine
density = N / D l
Feldman and Peters (1979) consider that the mag-
nitude of underestimation of spine density correlates
positively with dendritic shaft diameter and nega-
tively with spine length and they include these par-
ameters in their formula. Their equation (Eqn 2) is
based on several assumptions including the consist-
ency of the dendritic shaft diameter, the concentricity
of the area into which spines project, that spines are
at least as long as the spine head diameter and that
they are randomly distributed over the entire den-
dritic surface. The assumption of random distri-
bution of spines was tested and found to be
acceptable regardless of dendritic diameter size (Pe-
ters and Walsh, 1972), although shaft diameter has
been correlated with the density of visible spines
(Feldman, 1976; Feldman and Dowd, 1975). When
measuring spine density on the same designated
segment of dendrite in a specific dendritic type, e.g.
a 50 pm length of basal dendrite starting 20 #m from
its origin, there should not be a problem with this
equation. However, in a recent study (Homer and
Arbuthnott, 1991) ratios of spine density estimates
recorded using visible spine numbers and the
equation were compared between dendritic types. If
the spine distribution is random throughout the
dendritic field of a particular cell type then the
equation should result in proportionately greater
spine densities in dendrites of greater diameter i.e.
apical spine density should be much greater than
basal. Although spine densities recorded using the
equation were substantially larger than estimates
using visible spine counts, and apical dendrites did
have the greatest spine density when corrected using
the formula (Feldman and Peters, 1979), the magni-
tude of the difference between dendritic types was not
as expected when dendritic diameter was taken into
consideration (Horner and Arbuthnott, 199t). This
does not agree entirely with Feldman (1976) who
records spine densities which correlate positively with
dendritic shaft diameter.
The equation devised by Feldman and Peters
(1979) gives spine density results accurate to within
10% of the actual number which they corroborated
by testing on dendrites where the absolute spine
 PLASTICITY OF THE DmCOPaT]C SpX~rE
number was known. This relatively-accurate method
of assessing spine density is more time-consuming
than the counting of spines only, but is performed
relatively quickly with the use of semi-automatic
image analysis. It is not free of problems in that the
spines in visible zones may not be truly representative
of the population, and the dendritic shaft diameter
may not be the true average if the shaft is at all
flattened in the area measured. Due to differences in
judgement between observers, it is necessary that
whatever method is used the same observer should
perform all measurements to ensure that consistent
criteria are applied.
Other attempts at estimating the total number of
dendritic spines were made by Chan-Palay et al.
(1974) and Bradley and Berry (1976). Cban-Palay
et al. (1974) made estimates of spine density on
Meynert cells in the primary visual cortex. In an
attempt to estimate the actual number of spines, they
also assumed that they were distributed uniformly
over the surface of that segment of dendrite
measured. They claim that half the observed number
of spines per 100/am unit length multiplied by the
number of 100/zm units in the total dendritic length
assessed would give the linear number of spines in
any particular zone. The linear number calculated
multiplied by the fraction of the circumference of
dendrite to the distance between spines would give
the total number of spines (Eqn 3) in any particular
dendritic zone. Chan-Palay et al. (1974) did not take
into account the spine length, an important factor,
which appears to be extremely variable both between
different spine shapes and neuronal types. This cor-
rection formula also presents difficulties in appli-
cation to various cell types. If the distance between
dendritic spines is greater than the circumference of
the dendrite the formula is not applicable and poses
difficulties with small diameter dendrites with widely
spaced spines.
Equation 3 (Chan-Palay et al., 1974)
Linear number of spines = n/2
(number of 100 #m units in the total DI)
N = Linear number of spines
(Dc/distance between spines)
where N = estimate of the total number of spines,
n =visible number of spines per 100/zm seg-
ment of dendrite, DI-~ dendritic length assessed,
Dc = dendritic circumference.
In studies on deafferentation (Bradley and Berry,
1976) and undernutrition (McConnell and Berry,
1978) a correction formula (Eqn 4), to try to over-
come the inadequacies of visible spine counts, was
used. They too presumed an even distribution of
spines over the entire dendritic surface and hence the
greater the diameter of the dendrite, the smaller the
proportion of the total spines visible. The correction
factor takes into account the radius of dendrite and
the mean spine length which when formula gives an
estimate of the total number of spines per unit area
of the entire surface. However, the rationale for, or
297
the derivation of, the correction factor used is not
explained nor the evidence of its validity proven.
Equation 4 (McConnell and Berry, 1978)
Correction factor = (2r x arc cos(r/r + ))- i.
Total spine counts can be made directly although
these are difficult as routine procedures. They were
used by Feldman and Peters (1979) as methods of
validating their formula. One method is that of
reconstruction of the dendrites from serial thin sec-
tions oriented in a plane transverse to the dendrite
axis. Obviously, limited numbers of dendrites and
abbreviated lengths of dendrites can be analysed in
this way. It is also difficult to specify unequivocally
the identity of the dendritic segment. A second
method of performing total counts is by light
microscopy of longitudinally orientated Golgi-
impregnated dendrites which have been subjected
to deimpregnation (Fairtn et al., 1977). This method
allows assessment of greater numbers and longer
lengths of dendrites but requires effort to obtain a
sufficient number of suitably deimpregnated
examples (Feldman and Peters, 1979). For the pre-
sent, the most reliable, tested method of estimating
total dendritic spine numbers with reasonable
efficiency, in relatively large samples is by the formula
devised by Feldman and Peters (1979).
4. PLASTICITY OF THE DENDRITIC SPINE
4.1. INTRODUCTION
Plasticity is defined as "the capability of being
formed or moulded". Lynch et al. (1978) suggest that
in neuroplasticity the cells demonstrate the capability
for persistent change or adjustment. It is unlikely that
there is a single major mechanism responsible for
plasticity, but rather that it results from a number of
mechanisms (Carlin and Siekevitz, 1983). Functional
and morphological changes resulting in modification
of the response of a stimulated pathway by preceding
activity are likely to underlie neuronal plasticity
(Fifkov/t and Van Harreveld, 1977).
Dendritic spines may alter in size, shape or internal
structure. Changes in dendritic spine density are
important in that the dendritic spines possess the vast
majority of synaptic contacts, and therefore form the
main electrical circuits. The density of spines on
neuronal dendrites can increase or decrease depend-
ing on the experimental conditions. Early findings on
plasticity of dendrites and their spines are noted by
Monti (1895), Demoor (1898) and Querton (1898).
More recently, many studies have confirmed the
remarkable plasticity of spines. Very rapid (mins)
alterations in the morphology of dendritic spines
have been observed with behavioural manipulations
in the honey bee (Brandon and Coss, 1982) and the
jewel fish (Burgess and Coss, 1983), In the developing
chick brain, visual experience results in changes in
dendritic spines that correlate with increased neur-
onal excitability in the same brain region (Bradley
and Horn, 1979; Brown and Horn, 1979). In the
hippocampal formation, electrical stimulation of
298 C.H. HORNER
afferent pathways modulates dendritic spine mor-
phology and correlates with long-lasting changes in
neuronal physiology (Fifkov~i and Van Harreveld,
1977; Chang and Greenough, 1984). Studies which
have associated changes in dendritic spines with
experience and/or altered neuronal physiology, have
led to the repeated suggestion that dendritic spines
are a site of plasticity that could be involved in
processes of learning and memory (Diamond et al..
1971; Crick, 1982; Patel and Stewart, 1988).
Plasticity has been demonstrated in many regions
of the brain and probably occurs throughout the
entire central nervous system, the extent varying with
the experimental factor being tested. Commonly
studied sites include the cerebellum (Pentney, 1982;
Chen and Hillman, 1980; Sotelo et al., 1975), the
cerebral cortex, particularly the visual (Valverde,
1967; Parnavelas et al., 1973 and Uylings et al., 1978)
and auditory (Ryugo et al., 1975) sensory cortices.
Many studies have used the hippocampus which is
regarded as a suitable model because its circuitries
possess a remarkable capacity for both functional
and structural change (Alger and Teyler, 1976; Lynch
et aL, 1978; McWilliams and Lynch, 1978; Lee et al.,
1980; Chang and Greenough, 1984). In adult mam-
malian hippocampal formation both physiological
(Bliss and Lomo, 1973; Diamond et al., 1989) and
morphological (Desmond and Levy, 1988; Gould
et al., 1990b) changes have been observed in neurons
in response to a variety of experimental manipula-
tions and naturally-occurring phenomena. This
plasticity is particularly intriguing in that the hippo-
campus has been shown to be involved in cognitive
processes such as learning and memory (Olton, 1983;
Squire, 1983), which undoubtedly stem from changes
at the cellular level (Woolley et al., 1990). In so far
as neuronal structure has been shown to influence
neuronal physiology (Kawato et al., 1984) it seems
highly likely that morphological changes observed in
the hippocampus have physiological and ultimately,
functional consequences (Woolley et al., 1990).
Animal models are used, not only for ethical
reasons, but because their environment may be con-
trolled to a greater extent. However, evidence of
neural plasticity in the human exists in conditions
such as chronic alcoholism (Ferrer et al., 1986), senile
dementia and Alzheimer's disease (Scheibel et aL,
1975; El Hachimi and Foncin, 1990), temporal lobe
epilepsy (Scheibel et al., 1974), chromosomal abnor-
malities (Marin-Padilla, 1972, 1974) and mental re-
tardation states (Purpura, 1974).
Plasticity of the dendritic spine has been noted
both in neonates and adults (Chen and Hillman,
1980, 1982a,b; Gray, 1982), under experimental ma-
nipulation (Chen and Hillman, 1980; Fifkov~ and
Van Harreveld, 1977) and surgical disconnection of
their afferents (Chen and Hiilman, 1982a,b), and
implicated in learning (Crick, 1982; Fifkov~i and Van
Harrefeld, 1977; Gray, 1982) and several other func-
tions (Swindale, 1981). An example of their plastic
nature in the adult brain is shown by their behaviour
following axon lesioning. The synaptic terminal on
the dendritic spine degenerates and is phagocytosed
by astrocytes with part of the spine head (Colonnier,
1964), but the spine stump can regrow and make
synapses with adventitious axons (Gray, 1982).
Plasticity is even evident in transplanted neural tissue.
Zemanick et al. (1987) assessed spine density in 35
day old neostriatal transplants. Transplanted neurons
had decreased spine densities and numbers of proxi-
mal dendrites. It may be that the differences are due
to aberrant maturation of transplanted neurons~
4.2. GENDEREFFECTS
The structure of the brain is, in part, a function of
the environment in which an animal has been raised.
Rats raised in a complex environment have been
noted to have higher dendritic spine density (Globus
et al., 1973) and longer synapses (MoUgaard et al.,
1971) than those reared in socially deprived or iso-
lated environments. However, there appears to be
sexual dimorphism in the extent of plasticity found.
Juraska (1984) noted a greater male vulnerability to
the environment during development and suggested
that the neuronal response to experience might be
sexually dimorphic. The female visual cortex is not
simply a smaller version of the male, but has a
different pattern and the degree of difference is depen-
dent on the environment. In further studies in hippo-
campal (CA3) tissue, Juraska et al. (1989) noted sex
differences within the dendritic tree of neurons and
between the neuronal type which also responded
differently to the particular rearing environment. The
differences between the sexes were more prominent in
those from the enriched or complex environment. The
mechanisms for the occurence of such sex differences
and the functional significance of them are not yet
understood but are not simply due to hormonal
differences (Juraska et al., 1989). Although these sex
differences reported by Juraska et al. (1989) were
based on dendritic tree measurements, Gould et al.
(1990a) noted sex differences in both number of
primary dendrites and spines on apical dendrites of
hippocampal (CA3) cells in rats. Females had
more primary dendrites and males had more apical
excrescences.
Male-female differences have been recorded in
other areas of the central nervous system--the medial
amygdala--following lesioning of the posterior corti-
cal nucleus (Nishizuka and Arai, 1983). Post-lesion-
ing the number of degenerating axon terminals in the
ventral molecular layer was statistically greater in
males compared with females. Later (10 days), intact
synapses on spines of the ventral molecular layer were
significantly reduced in males compared with con-
trois, while females were not significantly altered. The
number of axon terminals in the male, originating
from the lesioned area was greater than in the female
and the number of synapses in the molecular layer of
the unoperated male was statistically greater than
that in the female. The sex differences in synaptic
number became undetectable 10 days after operation.
Sex differences also occur during the period of
development. In pyramidal cells of the visual cortex
the number of spines increased from 10-20 days
postnatally in both sexes but in females where the
spine numbers were already significantly greater than
males the absolute difference increased with increas-
ing age until a maximum occurred at 20 days. The
number of spines then decreased in females between
20 and 60 days, while they increased in males until
 PLASTICITY OF THE DENDRITICSPINE
at 60 days the numbers were similar in both sexes
(Munoz-Cueto et al., 1990). In cases where ovari-
ectomy had been performed on females at 30 days,
the decrease in spines was prevented, suggesting that
spine development in rat visual cortex is dependent
on sex and that sex hormones may have a role in
pyramidal cell maturation.
4.3. SPINEMORPHOLOGY
The dendritic spine may demonstrate its plasticity
either purely in terms of its structureor alterationin
density of spines.In 191 I, Cajal noted that soporifics
and other drugs had effectson the morphology of
spines. The spines along the apical dendrites of
cortical cells represent a sequence of postsynaptic
structures highly sensitive to any damage to
(Colonnier, 1964; Globus and Scbeibel, 1967b, c;
Valverde, 1968), or the functional state (Valverde,
1967), of their afferent fibres. Electron microscopy
study of synapses (axosomatic and axodendritic)
following extreme exposures to phenamin, alcohol
intoxication, local tetanus, shock and radiation
trauma revealed destructivechanges to synapses and
spinesmanifested by dissociationof pre- and postsyn-
aptic parts and disorganisationof the spine apparatus
(Turmanov, 1984). Malnutrition has produced giant
spines and elongated synapses (Chen and Hillman,
1980).
In normal physiologicalstatessuch as hibernation,
Querton (1898) noted different dimensions in den-
dritic spines between hibernating and awake alpine
marmots. Confirmation of morphological changes in
spines between hibernating and awake states in
ground squirrels (citellis lateralis, citellis tridecemlin-
eatus) were noted by Boycott (1982). The postsyn-
aptic knobs were larger in hibernators and though the
necks of spines were the same length they were thicker
than in awake animals. The maximum diameter of
knobs was 1 pm in hibernators and < 0.5 #m in
awake animals. The complete changes from a hiber-
nating to an awake spine occurred within 12-24 hr.
In material from aged rats, a fragmented and
darkened spine apparatus in spines is seen occasion-
ally (Feldman, 1976). It is not known whether these
features are related to age but one such feature does
appear to represent dendritic membranous inclusions
in the shaft. These inclusions ( > 0 . 5 # m and
< 1.0#m in diameter) are heteromorphic and
increase in frequency with age. The significance of the
inclusions is not clear but they may be related to
lost spines (Feldman, 1976). Many other examples of
changes in spine morphology exist but the structural
changes are often concurrent with fluctuations in the
spine density.
4.4. SPINEDENSITY
Dendritic spine density is not a constant feature
between or within cell types, the density varies be-
tween locations of the dendritic field--the dentate
gyrus cells display greater variation along the den-
drite than other limbic neurons. Different cell types
also vary greatly in their density, cortical pyramidal
ceils being particularly spiny while nonpyramidal
cells are less so. Spine density of various neuronal
J]PN 4L/3~B
299
types--pyramidal, Purkinje, stellate---increase and
decrease with different experimental conditions and
procedures (Parnavelas et al., 1973; Ryugo et al.,
1975) but normal aging (Feldman and Dowd, 1975;
Feldman, 1976) and gender (Juraska et al., 1989;
Nishizuka and Arai, 1983; Munoz-Cueto et al., 1990)
(Section 4.2) also play a part in the response.
Towards the end of the 19th Century several
conditions including starvation, freezing, shock and a
variety of drugs were known to lead to a loss of spines
(Monti, 1895; Demoor, 1898; Querton, 1898). These
researchers explained their results using a "theory of
protoplasmic plasticity" of neurons based on the
notion that physiological stimulation of neurons was
accompanied by retraction of its spines and dendrites.
Since then a vast number of studies have confirmed
the plasticity of the dendritic spine in terms of
density.
Why do spine numbers change? Is it a direct result
of the experimental condition? Are increases in spine
density in one area a compensatory mechanism for
deficits due to deprivation of sensory input, e.g.
iesioning of an afferent pathway, or do they reflect
increased synaptic contacts to meet a demand for
integration of greater sensory input? Many studies
have demonstrated that dendritic spines are especially
sensitive to sensory deprivation. A decrease in the
number of dendritic spines occurs following sensory
deprivation, caused either by modifications of the
normal environment (Globus and Scheibel, 1967d;
Valverde, 1967) or by manipulation of the sensory
pathways (Colonnier, 1964; Globus and Scheibel
1967b,c; Valverde and Esttban, 1968). In broad terms
the spine losses attributable to aging can be seen to
be of the same magnitude as those produced in young
animals in experimental conditions. Feldman (1976)
queries the temporal relationship between the loss
of a spine and the loss of the synapse on that spine
during the aging process. He claims that it is
unknown whether the loss of a spine is a transneu-
ronally-induced occurrence--an intrinsic phenom-
enon that secondarily leads to degeneration of the
axon terminal--or whether the two processes are
initiated by unrelated events. Since deafferentation
results in a reduction of pyramidal dendritic spines
(Gruner et al., 1974; Globus and Scheibel, 1967b, c;
Valverde and Est~ban, 1968; Ben Hamida et al., 1970)
it seems likely that spine loss is transneuronally
induced. Valverde (1967) maintains that the spine loss
following sensory deprivation is due to experience
deprivation while Scheibel et al. (1974) suggest that
degenerative processes or transynaptic degeneration
produce loss of dendritic spines.
Conversely, increased spine densities have been
demonstrated in rats reared under "enriched" en-
vironmental conditions (Globus et al., 1973; Schapiro
and Vukovich, 1970). Electron microscopic studies
have emphasised the importance of such observations
in showing that dendritic spines are postsynaptic
specialisations and major sites of synaptic input to
cortical pyramidal cells (Colonnier, 1964; Gray,
1959a,b). Current evidence suggests that spine density
changes may reflect modifications in synaptic input.
If it is assumed that synaptic inputs of different types
are associated with specific portions of the dendritic
tree (Globus and Scheibel, 1967a, b,c; Valverde,
300 C.H. HORNER
1968), then synaptic loss is not limited to any one
afferent system.
A more detailed report of the plasticity of spine
morphology and density in both naturally occurring
phenomena such as aging and hormonal cycles and in
experimentally induced states such as deafferentation
or stimulation of pathways as well as in certain
pathological states, now follows.
5. NATURALLY-OCCURRING PLASTICITY
5.1. INTRODUCTION
The dendritic spine appears to be sensitive to
naturally-occurring phenomena as well as experimen-
tally-induced conditions. With the passage of time
from the embryonic period to old age (Section 5.2)
the spine density fluctuates, The development and
maturation of the central nervous system is also
affected by hormones, particularly thyroid and go-
nadal. Physiological fluctations of these hormones
are reflected in changes in spine density (Section 5.3).
Other naturally-occurring conditions such as hiber-
nation (Querton, 1898; Boycott, 1982; Sanchez-
Toscano et al., 1989) and temperature change
(Boycott, 1982) have concurrent changes in synaptic
function through altered structure of synapses on
spines and changes in spine density. Therefore, it
seems that spine density reflects the functional status
of neurons.
5.2. TEMPORALVARIATION
The dendritic spines and their synapses appear to
change continually during life even in regions with
highly stable neuronal populations (Bondareff and
Geinisman, 1976). Lifelong change in spine density is
consistent with the view that the adult brain readily
demonstrates plasticity especially under experimental
conditions. The continuous change in spine number
also indicate that caution is necessary in defining
"normal" or "adult" spine densities. During develop-
ment the spine complement increases but an adult
level is reached soon after birth. Spine development
is thought to be an essential part of the physiological
maturation of the brain (Ruiz-Marcos and Valverde,
1969; Juraska and Fifkowi, 1979) and the spine
density varies significantly as a function of age
(Murphy and Magness, 1984). In the first stages of
development the growth and maturation of the ner-
vous system are influenced closely by thyroid hor-
mones. Deficiencies at this stage can lead to decreased
numbers of synaptic spines (Mussa et al., 1989).
Subsequently the process of normal aging gradually
reduces the dendritic spine density (Bondareff and
Genisman, 1976; Feldman, 1976; Feldman and
Dowd, 1975).
5.2.1. Development
Spines appear slowly during development in associ-
ation with the increasingly complex behavioural and
neurophysiological changes (Scbeibel and Scheibei,
1968). In the study of the ontogenetic development of
dendritic spines, Minkwitz and Holtz (1975) used
dendritic spine estimations as an indicator of
neuronal change and development in hippocampal
pyramidal cells. In the early postnatal period, post..
synaptic structures (spines) appeared in small bu~
increasing numbers but the initial segment of the
apical dendrite remained free of spines, At 20 days
postnatally, following the spine-free initial zone there
was a rapid increase in spine numbers reaching a
maximum approximately midway along ~hc ~pical
dendrite which then slowly decreased. At rials .~tage,
the basal dendrite acquired its adult complement of
spines while the oblique or lateral dendrites were
expected to increase further as were the apical end
tufts which however did not reach the values of other
dendrites. In the period of development and aging
from the 21 day embryo to the 20 day animal the
spine numbers increase from 20 to 3000 which is
equivalent to 0.02~).55 spines per m~crometer
(Minkwitz, 1976).
In all studies of the mammalian visual cortex there
is an increase in spine density during the early
postnatal period (Boothe et al., 1979; Juraska and
Fifkovfi, 1979), although the rate and pattern of
increase of spines is not uniform across all branches.
Secondary and tertiary branches peak later and the
peak is followed by a plateau but more often a
significant decrease to adult levels (Murphy and
Magness, 1984). Study of the somatosensory (Galofre
and Ferrer, 1987) and sensorimotor (Petit et al., 1988)
cortices showed similar trends with dramatic in-
creases in spine density at an early stage followed by
a decrease, mainly on proximal dendrites; the density
continuing to increase on terminal branchesl The time
period during which spine density peaks above adult
levels may coincide with and provide a morphological
correlate of, the critical period (Murphy and
Magness, 1984).
In human neocortex the number of dendritic spines
on apical dendrites has been shown to increase from
460 in an 8 month foetus, to 498 in a premature
infant, to 1412 in a newborn infant. The increase
continues to 1874 in a 45 day old and 2100 in a 60
day old baby (Marin-Padilla, 1967). Michel and
Garey (1984) also studied spine development in hu-
man visual cortex from the 33rd foetal week to 30
years postnatally. They recorded spine densities of
30/50#m in late foetal life and 50/50/~m in the
neonate. By 5 months old a maximum spine density
of approximately 80/50/~m was attained alter which
spine numbers decreased back to 50/50 #m by 2 years
and maintained this level into adulthoood
Morphological modifications were also noted
during development. The mushroom-shaped, stubby
and short-thin spines in one month old rats replaced
long thin pedicled spines of earlier development.
Between 3 and 6 months there was variability in
density and individual variability with regard to the
percentage of different types of spine on apical den-
drites among cells in the same animal (Galofre and
Ferrer, 1987). More detailed study of the dendritic
spine with age has revealed that the calcium distri-
bution in dendritic spines of dentate granule cells
varies as a function of age (Fifkov~i and Cullen-
Dockstader, 1986). Postnatally, the percentage of
spines with a spine-apparatus containing calcium
precipitates decreased significantly. Conversely the
percentage of spines with precipitates in the spine
 PLASTICITY OF THE DENDRITIC SPINE
cytoplasm increased significantly. In the absence of a
spine-apparatus loss, this suggests an age-related
decrease in the calcium-sequestering capacity by the
spine apparatus. The parameters improved by the
thirtieth postnatal day and approximated the early
values (day 3).
The spine formation and maturation of type I
synapses has been studied on several cell types in the
visual cortex of rabbits (Muller et al., 1984), macaque
monkeys (Mates and Lund, 1983) and cats
(Kakabadze, 1985). On spiny stellate cells of the
striate visual cortex, Mates and Lund (1983) noted
that 95% of type I synaptic contacts were originally
found on dendritic shafts, and appeared to be carried
out on spine outgrowths. In the adult, the type I
contacts were only found on spine tips. During later
maturation, the spine bearing type I contacts may be
lost, since profiles of the neuropil illustrated shrink-
age and detachment of type I axon terminals and
postsynaptic dendritic spines in normal maturation,
which may explain the marked increase and decrease
in spine population on these neurons during normal
maturation noted by Boothe et al. (1979). In the early
postnatal period, synapses on pyramidal cells were
occasionally immature in appearance, characterised
by poorly defined postsynaptic densities that were not
easily defined as type I or II (Muller et al., 1984).
Distinction between types became easier and in adults
the postsynaptic density of asymmetrical (type I)
synapses was much thicker and spines were found to
have only type 1 synapses which increased in fre-
quency. Kakabadze (1985) also recorded that asym-
metric (type I) synapses appeared at a later stage in
cat visual cortex.
In developing motor cortex in the monkey, the first
synapse was noted on day 53 of the embryonic period
(Zecevic et al., 1989). By midgestation (day 89) there
were synapses observed over the entire width of the
cortical plate at a density of 5/100/~m 3. In the last 2
months gestation the density increased eight-fold to
40/100/tin3 at birth (day 165). The increase in density
continued postnatally until the end of month 2 when
it reached 60/100 #m 3, which is twice the adult level.
It then decreased gradually until sexual maturity at
3 years and then more rapidly to adult level of
30/100/zm 3, when it remained relatively stable. The
decline after the peak in infancy was predominantly
at the expense of asymmetric synapses. Therefore the
initial overproduction was followed by selective elim-
ination and structural alterations.
5.2.2. Aging
The distribution and density of synapses is of direct
functional significance in aging. While Feldman and
Dowd (1974) recorded losses of dendritic spines,
Scheibel et al, (1975) noted a reduction in dendritic
arborisation in aged cortex. Age-related spine loss
affects the entire neuron but not necessarily all neur-
ons. No selective depletion of any particular spine
type occurs and the variability in loss is not under-
stood. Spine losses ranging from 24% on terminal
tufts to 40% on oblique branches on pyramidal cells
of rat pyramidal cortex were noted between three
month and three year old animals, with an average of
one third of all spines lost (Feldman, 1976). Leuba
301
(1983) recorded a 50% decrease in dendritic spines
between the ages of six months and two years in mice,
but most of the spine loss occurred by one and a half
years. Hippoeampal (CA1) pyramidal cells had sig-
nificantly reduced spine densities of approximately 12
and 10.5% on basal and apical dendrites respectively
in aged (27 month) animals (Nunzi et al., 1989). The
cortical pyramidal cell spine losses due to aging were
at least as severe in visual cortex of the rat as those
due to sensory deprivation (Feldman and Dowd,
1975). This holds true for comparisons of spine loss
in other species and cortices (Valverde and Estrban,
1968; Valverde, 1967; Globus and Scheibel, 1967b, c)
although undercut cat cortex produced significantly
greater spine reductions (Rutledge et al, 1972). The
main correlate of variability among apical dendritic
spine density was the dendritic diameter (Feldman,
1976). Feldman (1976) found that thicker dendrites
have greater spine densities than thinner dendrites
and the relative spine loss with age differed between
thick and thin apical dendrites, being 20% for thick,
36% for medium and 31% for thin diameter
dendrites.
Since spines are postsynaptic specialisations, neur-
ons with spine depletion also probably undergo alter-
ations of synaptic complement. Synapses may
disappear or new synapses might form on remaining
spines on dendritic shafts. Cragg (1975) failed to
demonstrate any age-related change in the number of
neocortical synapses in human brain but Bondareff
and Geinisman (1976) disagree having recorded a
27% decrease in synapses on molecular cells of the
dentate gyrus in 25 month old compared to 3 month
old rats. The actual loss of synapses involved a 26%
decrease in those on spines which was less than that
on shafts in senescent animals. Feldman (1976) also
recorded a loss of 20% of total synapses on shafts
and spines in rat visual cortex pyramids over a 3 year
period, the synapses on spines decreasing by 26%
which produced a relative shift in the proportions of
spine and shaft synapses with age. There was no
evidence of increased numbers of synapses on surviv-
ing spines. A postnatal study of synapses on hippo-
campal cells in rats showed no significant difference
in the density of synapses and postsynaptic dendritic
spines between three and twenty four months (Lolova
et al., 1989). However the area of the presynaptic
terminals and the postsynaptic spines was decreased
significantly and the overall size of the synaptic
contact zones was decreased in the older animals. The
synaptic density may not vary between these two age
groups because the older group are passed the peak
stage and the numbers have reverted to levels similar
to those found prior to the peak. Since degenerative
processes usually lead to regeneration or axonal
sprouting (Lynch et al., 1978) loss of synapses in
senescent brain suggests a relationship to regener-
ation phenomena which determines the functional
recovery of the central nervous system after lesions
(Bondareff and Geinisman, 1976).
The charhcteristic alterations in spine density with
age may correlate with other factors determining the
plasticity of an individual neuron's functional deter-
ruination (Murphy and Magness, 1984). Age not only
affects the density and morphology of spines and
their synaptic contacts but also affects the response of
302 C.H. HORNER
these structures to other experimental conditions.
Vaughan and Cahill (1984) sectioned the corpus
callosum of rats aged 3, 12 and 24 months and
examined the auditory cortex. They found age-
related differences in the morphological response
to partial deafferentation. Normally the apical den-
drites lose spines and become thinner with age
(Feldman, 1976). In lesioned animals the apical
spine density decreased in one layer (III) while it
increased in another (IV). The spine densities varied
with age, being highest in layer III apical dendrites in
older animals and highest in layer IV in the younger
rats.
5.3. HORMONALVARIATION
5.3.l. Thyroid
Thyroid hormones play an important rote in nor-
mal development and maturation of the nervous
system and in particular the cortical pyramids and
Purkinje cells. Purkinje cell maturation in the frog
cerebellum during thyroxine-induced metamorphosis
was studied by Hauser and Gona (1984). Following
treatment of premetamorphic bullfrog tadpoles with
thyroxine, both populations of Purkinje cells showed
rapid maturational change and developing Purkinje
cell spines formed synapses with parallel fibres. The
dramatic nature of thyroxine action in inducing the
complex series of detailed maturational changes in
the cerebellum was 1-2 years ahead of schedule.
Thyroxine-induced Purkinje cell maturation is more
rapid than that seen during spontaneous metamor-
phosis and therefore maturation during metamor-
phosis is largely dependent on thyroxine. Gould et al.
(1990a) found that neonatal treatment with thyroid
hormone produced long lasting and dramatic
changes of CA3 cells in adults which were sexually
dimorphic. The thyroid treatment significantly in-
creased the numbers of primary dendrites and apical
excrescences but the treatment differences occurred
in both sexes and were of equal magnitude regardless
of sex. However the hippocampal region (CA1)
did not show any sex differences with or without
treatment.
Hypothyroidism results in decreased numbers of
synaptic spines and internuncial connections (Mussa
et al., 1989): the maturation process is incomplete,
resulting in impairment in content and organisation
of the intra-cytoplasmic microtubulin and biochemi-
cal maturation of synaptosomes.
5.3.2. Gonadal
Gonadal steroids are known to influence hippo-
campal physiology in adulthood. Woolley et al.
(1990) studied the structural mechanisms that may
underlie hippocampal function in the normal adult
brain by identification of the endogenous factors that
mediate naturally occurring, as opposed to exper-
imentally induced, neuronal plasticity. "l'he pattern of
change in dendritic spine density of hippocarnpal
(CA1) cells during the oestrus cycle correlates well
with the changing levels of oestradiol and progester-
one (Smith et aL, 1975; Wooiley et al., 1990). During
proestrus, when oestradiol and progesterone are at
their highest levels, dendritic spine density values are
also greatest. Intermediate oestradiol levels during
dioestrus correlate with the intermediate spine den-
sity values and the lowest hormone levels in the
oestrus accompany the lowest spine density. How-
ever the significant differences observed in apical
dendritic spine density were not found in the basal
tree of these cells but a trend towards a decrease was
evident in the oestrus phase of the cycle compared to
the proestrus phase. Also, the changes noted in CAI
cells were not found in either CA3 pyramids or
dentate granule cells. These results demonstrate nat-
urally occuring structural plasticity in a specific
population of neurons in the adult mammalian hip-
pocampal formation. Fluctuation of the spine den-
sity during the oestrus cycle was confirmed in a study
of VMN (hypothalamic) neurons in rats (Frankfurt
et al., 1990). That dendritic spines are remarkably
plastic is shown by a decrease of over 30% during
approximately 24hr between the proestrus and
oestrus (Smith et al., 1975). It then fluctuates from
low values during oestrus to high values in proestrus
within a period of days. Longer-term differences in
hormone levels may result in a greater degree of
morphologic variation. On-going morphologic plas-
ticity has been observed within the CA1 hippocam-
pal pyramids, where changes in apical spine density
have been observed in developing male mice during
puberty and with castration (Meyer et al., 1978).
In ovariectomized rats, a profound decrease in
dendritic spine density of CAI hippocampal pyra-
mids was noted when compared with those on re-
placement treatment with oestradiol or oestradiol
and progesterone (Gould et at., 1990). Again
although the CA1 area was affected, CA3 showed no
change with ovariectomy or replacement therapy.
Adult rats who had undergone ovariectomy had
significantly reduced spine densities in hypothalamic
neurons when compared with intact females or those
who had received oestrogen or oestrogen plus pro-
gesterone therapy (Franfurt et al., 1990).
The modulation of dendritic spine density may be
a direct effect of ovarian steroids on CA1 pyramids,
in that Loy et al. (1988) reported that pyramidal cells
of CA1 accumulate oetradiol while only a few CA3
pyramids and no granule cells do, and oestrogen-in-
ducible progesterone receptors have been reported in
the CA1 region (Parsons et al., 1982). However since
CA I apical and basal dendrites are differentially
sensitive to variations in the levels of ovarian hor-
mones, there must be an inherent difference in their
morphologic plasticity or an indirect factor involved
(Woolley et al., 1990).
Change in spine density due to fluctuating ovarian
steroid levels may result in alteration in number
and/or organisation of excitatory synapses and
thereby modulate the excitatory input to CA1 pyra-
midal neurons. Oestradiol appears to regulate
synapse density in the stratum-lacunosum molecu-
lare of the CA1 region (Woolley and McEwen,
1990). If increased dendritic spine density does reflect
increased excitatory input, which could, in turn,
increase neuronal activity, then the results of Wool-
Icy et aL (1990) could present a possible structural
mechanism to explain the effects of ovarian steroids
on hippocampal neuronal excitatibility.
 PLASTICITY OF THE DENDRITIC SPINE
6. EXPERIMENTALLY-INDUCED PLASTICITY
6.1. INTRODUCTION
The morphology and density of dendritic spines in
the cerebral cortex reflect environmental conditions,
the sensory systems of animals being sensitive to
experimental manipulations. These changes may in
turn reflect modifications in synaptic input (Ryugo
et al., 1975). Multiple experiments have tested the
plasticity of the dendritic spine to varying sensory
information. Both invasive and noninvasive pro-
cedures have been used. In terms of invasive pro-
cedures, pathways have been stimulated electrically
and sensory input has been interrupted by
deafferentation. Noninvasive experiments have em-
ployed varying degrees of sensory stimuli through the
environment in which the animals are maintained i.e.
either deprived of sensory stimuli such as light or
stimulated by a variety of sensory stimuli using a
complex environment.
6.2. DEPRIVATION
Several experiments testing the effects of depri-
vation confirm the significant loss of dendritic spines.
In monkeys deprived of somatosensory-motor ex-
perience, apical dendritic spine densities of primary
motor and somatosensory cortices were reduced but
visual cortex was spared since these animals had
visual access of the colony. The control colony and
a second group who had visual but no tactile access
to the colony, and who had cage equipment which
provided tactile stimuli, produced similar results
(Bryan and Riesen, 1989). In addition, as other
dendritic parameters were unchanged the measure-
ment of spine density seems to be a more direct
measure of early environmental deprivation.
6.2.1. Deafferentation
Dendritic spines disappear when the synaptic ter-
minals impinging on them are destroyed. Terminal
synaptic sites of axonal projections can be mapped by
interrupting the afferent system. Globus and Seheibel
(1967c) recorded a loss of dendritic spines (approxi-
mately 30%) limited to the oblique branches of apical
dendrites in the parietal cortex of 30 day old rabbits
after sectioning of the corpus callosum at birth. The
limitation of the changes in the spine count to the
oblique branches suggested that the callosal afferent
system, coming largely from the homotopic locus in
the contralateral cortex terminated on these branches
and since the spine loss was partial other afferent
systems must also synapse along these oblique
branches. Colonnier (1964) also recorded the disap-
pearance of spines following deafferentation, which
he reported to be resorbed when the presynaptic
influx is compromised or destroyed.
The concentration of spines upon apical dendrites
has been proven to be a reliable measure for dis-
tinguishing denervated from intact cortex (Rutledge,
1969). A quantitative study of the spine density of
cortical pyramids of the suprasylvian gyrus of the cat
following isolation showed a similar distribution of
spines in both isolated and nonisolated gyrus
303
(Benhamida et al., 1970), but a significant loss in the
number of spines from the basilar and apical den-
drites of pyramidal cells in each layer was noted.
However, a large number of spines remained and
showed intracortical synaptic connections. The loss
of the spines was attributed to the degeneration of
afferent fibres to the pyramidal cells which may be
retrograde or transynaptic.
Rutledge et aL (1972) confirmed that a significant
loss of dendritic spines occurs following denervation
of the cerebral cortex in both kittens and adult cats.
As young animals have not fully developed, the
effects of denervation upon spines would be very
important and should reflect alterations of develop-
ment and extent of persistence of synaptic contacts.
Dendritic spines are increasingly vulnerable to the
effects of denervation as the brain matures, the
maximum loss being 50% in adult undercut cortex.
The loss of spines in both young and mature animals
argues against the maintenance of functional connec-
tions if they are established following denervation at
any stage since axonal collateral proliferation did not
result in any maintenance of synaptic spine contacts
(Rutledge et al., 1972).
At the acute stage (36-48 hr following deafferenta-
tion of the cortex), the majority of dark degenerating
boutons still had contact with dendritic spines but at
the chronic stage (30--46 days postdeafferentation)
axon terminals decreased by 20--30% and the large
majority of boutons lost were those formerly synaps-
ing with dendritic spines. Therefore denervation leads
to a loss of afferents, loss of dendritic spines and
neuronal atrophy. A further feature was an increased
number of spine-apparatuses located in an extra-
spinous position in dendritic profiles (Gruner et al.,
1974). Following callosal deafferentation of the audi-
tory neocortex in the rat both callosal and thalamic
afferent synapse sites were examined (Vaughan and
Peters, 1985). Asymmetric synapses were distributed
with 84% on spines and 16% on shafts. Callosal
(93%) and thalamic (78%) afferents prefer to synapse
on spines. In a previous study Vaughan and Foundas
(1982) noted that 3 months after callosal lesioning of
1 month old animals, additional thalamic axons grow
and proliferate in the auditory cortex. In the long-
term callosally lesioned rats, thalamic axons did
exhibit specificity. After proliferation into the callosal
domain, 80% of degenerating terminals synapse with
dendritic spines and 20% with shafts which is like the
normal distribution of thalamic axons in this region.
Benshalom (1989) investigated the effects of neural
degeneration of presynaptic thalamo-cortical affer-
ents on dendritic spines of spiny stellate ceils in mouse
somatosensory cortex. The mean width of the spine
head increased significantly by 11% and the length of
the spine stalk by 25% in spines receiving thalamo-
cortical afferents. The mean spine stalk width at its
narrowest cross-section was 0.12 #m which was not
significantly different between these spines and those
with unidentified afferents. As structural features,
typifying transneuronal degeneration were not ob-
served along the spines examined, it was speculated
that morphological differences between spines may
result in part from degeneration-induced synaptic
inactivity of thalamo-cortical axospinous synapses
304 (. H. HORNER
rather than exclusively from any direct effects of the
degenerative process.
Ryugo et al. (1975) looked at changes occurring
within a sensory system that was an intact modality
following deafferentation of another sensory system.
After enucleation, no change was noted in spine
densities within the somatic sensory cortex but de-
creased numbers (30%) of spines on apical dendrites
in the rat visual cortex were found. Fifkovfi (1970)
found a similar decrease (28%) in a similar exper-
imental group following early lid suture. Alterna-
tively after vibrissae removal, Ryugo et al. (1975)
noted that the visual cortex remain unchanged but
decreased spine numbers were recorded in the so-
matosensory cortex. This suggests that spine losses
following deafferentation are restricted to the cortical
area of the affected sensory modality. Ryugo et al.
(1975) questioned what happened in the remaining
cortical areas following deafferentation of both visual
and somatosensory cortex and discovered a possible
compensatory increase in spine density in an intact
sensory area. Increased spine densities (22% in the
enucleated and 20% in those with vibrissae removal)
were demonstrated in the auditory cortex of these
animals following either form of deafferentation. This
apparent sensitivity of pyramidal cells of the auditory
cortex suggests an increase or redistribution of
synapses on auditory cortex cells following damage to
other sensory systems. The reason for this effect is
unknown. Ryugo et al. (1975) suggested that it may
be a response to some "trophic factor" released by
the deafferentated area or that the increased spine
incidence reflects increased use of the auditory sys-
tem. Therefore blinding or vibrissae removal may be
interpreted as a form of auditory "enrichment". In
these rats deprived of somatic or visual sensory input
via deafferentation the auditory system may assume
greater importance with increased utilisation of the
auditory modality. This increased use may be
reflected in spine density increments and in this sense
may resemble responses in rats reared in "enriched
environments" (Globus et al., 1973; Schapiro and
Vukovich, 1970) (Section 6.3.1).
Both enucleation and lesioning of the lateral gen-
iculate body unilaterally (interruption of the specific
sensory radiation to the visual cortex) in the newborn
rabbit leads to changes in dendritic spine numbers
only on apical shafts of pyramidal cells in the visual
cortex (Globus and Scheibel, 1967a). With enucle-
ation, the diminution in spines was 30% or more
while a more severe loss was noted with lesioning of
the lateral geniculate body ( > 65%), although the
distribution of loss was similar for both. The normal
appearance of other cell types in the visual cortex
indicated that the changes on the pyramidal cells were
direct rather than transneuronal which is the case
with enucleation. The sensitivity of postsynaptic com-
ponents to changes in level of presynaptic input
demonstrates the accuracy of the dendritic spine as an
index of presynaptic geometry providing an effective
measure of terminal axonal patterns throughout the
nervous system.
Valverde (1968) found a similar result in the striate
cortex. A significant decrease in the mean number of
spines in portions of apical dendritic shafts of pyra-
midal cells of the striate cortex receiving commissural
axons from the contralateral striate cortex affected by
enucleation have also been recorded (Valverde and
Estrban, 1968). Therefore, enucleation on one side,
affects the distribution of dendritic spines on the
apical shafts of pyramidal cells of the cortex on the
same side, presumably in response to the involvement
of the contralateral striate area.
Chemical destruction of the inferior olive in adult
rats (Sotelo et al., 1975) produced not only mainten-
ance of all spiny branches but also an unexpected
increase in the number of spines emerging from
secondary and tertiary thick Purkinje cell dendritic
trunks. The newly formed spines were already numer-
ous in rats which survived for 2 weeks and were still
present at 3 months post chemical injection. In this
study the fate of the Purkinje cell dendritic spines
contrasts with those reported by Hamori (1973). In
this chronic experiment the Purkinje cells were totally
deafferented from climbing fibres and some of the
new spines occupied abnormal positions on thick
dendritic trunks. Therefore, climbing fibres are not
necessary for the maintenance of dendritic spines but
on the contrary their absence induces spine for-
mation. The formation of new synaptic contacts
demonstrates the plasticity occurring in the reorgan-
isation of the cerebellar circuitry in abnormal con-
ditions (Sotelo, 1975; Sotelo and Arsenio-Nunes,
1976).
In a further study (Chen and Hillman, 1982a), a
marked plasticity of Purkinje cell dendrites in adult
animals occurred following a combination of granule
cell reduction and parallel fibre sectioning. Modified
Purkinje cell dendritic trees with high densities (5-fold
increase) of vacated spines on terminal branches as
well as on large dendrites were located in the thin
molecular layer. Ultrastructural analysis of synapses
on Purkinje cell spines (Chen and HiUman, 1982b)
revealed degeneration with vacating of postsynaptic
sites within 6 hr. Reactive synaptogenesis from re-
maining parallel fibres occurred even before degener-
ating parallel fibres had vacated postsynaptic sites. In
addition, new synapses formed between the shafts of
Purkinje cell branchlets and parallel fibres. Sprouting
of parallel fibres occurred as small extensions without
tubules while Purkinje cell spines reacted by forming
elongated and multiple heads which contacted differ-
ent parallel fibres. After 5 days, degenerating boutons
were rarely found. Enlarged spine heads were each
capped by a proportionally enlarged parallel fibre
bouton and joined by an elongated synaptic junction
to parallel fibres. Therefore, the formation of pre-
and postsynaptic sites take precedence over reoccupa-
tion of original contacts and multiple synapses on
individual spines are eliminated to give rise to single
contacts with boutons. This elimination results in
enlargement of synaptic contact areas between Purk-
inje cell spines and parallel fibres by taking over
postsynaptic sites from some vacated and eliminated
boutons.
In a study of the long-term effects of mossy fibre
degeneration in the rat, the primary response to
deafferented fibres was transneuronal degeneration of
the granule cell system, followed by secondary
transneuronal degeneration of Purkinje and molecu-
lar layers (Anderson and Flumerfelt, 1984). There
was an overall decrease in spine numbers on terminal
 PLASTICITYOF THE DENDRITICSPINE
branches and several existing spines increased their
length by as much as two-fold. When spines increased
their length, the Purkinje cells acquired new synaptic
contacts.
The effects of unilateral lesions of deep cerebellar
nuclei on the corticoeortical projection from the
somatosensory to the motor cortex in cats was stud-
ied (Keller et al., 1990). The number of corticocorticai
synapses/unit area in the experimental motor cortex
was significantly increased over controls. This was
due mainly to an increase of axon terminals synaps-
ing with spines belonging to pyramidal neurons. This
suggested that lesions of the cerebellar nuclei induce
sprouting of axon terminals in the motor cortex to
establish a new function. This is evidence for gener-
ation of new synapses in the adult CNS which is not
induced by elimination of existing synapses.
Denervation of some neurons may lead to pro-
gressive degeneration leading to cell death or in
others there may be compensatory new growth with
axonal proliferation. Whether proliferation occurs
depends upon such critical factors as neuronal age,
type and location of neurons and whether
deafferentation has been partial or complete
(Rutledge et al., 1972). Surgical removal of a collec-
tion of afferents and measuring the response of
residual inputs to the presence of newly denervated
dendritic space has been used to study possible
anatomical growth in the circuitries of the adult
hippocampus where it has been shown that the
remaining afferents undergo a growth response
("sprouting") and reinnervate the vacated territory
(Lee et al., 1977; Matthews et al., 1976b). Unilateral
removal of the hippocampus results in a 2 5 0 loss in
the terminal and synaptic population of the inner
molecular layer of the remaining, contralateral, den-
tate gyrus. Six weeks later, the deafferented zone
possesses an essentially normal population of termi-
nals and synaptic connections (McWilliams and
Lynch, 1978). In a study where the entorhinal cortex
has been removed, as much as 90% of the terminals
of the middle molecular layer are lost, but 5--6 days
after the lesion the terminal and synaptic population
is replaced, so that 5 days after sprouting has begun
nearly 20% of the normal synaptic bouton density is
recovered (Lee et al., 1977). There is strong evidence
that these "replacements" come from branches, emit-
ted from the intact axons found in the lower part of
the dendritic tree (Lynch et al., 1977), the growth of
the sprouting response being more vigorous in
younger animals. These experiments show that the
adult brain retains the ability to increase the number
and distribution of synaptic contacts formed by
specific afferents and there is evidence that the new
circuitries formed in this fashion are functional (West
et al., 1975; Tsukahara and Fujito, 1976).
Degenerating endings and terminal growth profiles
were found in the brains of"normal" rats (Sotelo and
Palay, 1971) which suggested that sprouting may be
a continuous process which strengthens the possi-
bility that physiological (long-term potentiation) and
behavioural change might be due to the use of the
capacity for growth in brain circuits (Lynch et al.,
1978). Reactive synaptogenesis through sprouting
has been demonstrated in the hippocampus
(Matthews et al., 1976a, b), septum (Field, et al. 1980;
305
Raisman and Field, 1973), optic system (Lund and
Lund, 1971) and cerebellum (Chen and HiUman,
1982b). This evidence shows that after partial
deafferentation, certain spared afferents can establish
synaptic connections on denervated neurons
(Cotman and Nadler, 1978). However, whether or
not these postsynaptic sites are reutilised is ques-
tioned. Raisman and Field (1973) suggested that
postsynaptic sites are permanent and are reoccupied
by sprouting afferents but Smolen (1981) claims that
contacts are made on entirely new synaptic sites.
Chen and Hillman (1982b) suggest that the mechan-
ism for the enlargement of synapses is the takeover of
vacated sites by newly-formed synapses as well as
synapse elimination. The ability of new synapses to
form in a very short time period, even before synaptic
sites are vacated, strongly supports a turnover
mechanism.
The rapid occurence of synaptogenesis indicates
that the postsynaptic element contains the potential
to form synapses with afferents which are in direct
contact. Postsynaptic structures can reorganise to
shift the position and size of synaptic connections,
possibly by returning to an immature-like state (Chen
and Hillman, 1982b). A second stage of postsynaptic
control over reactive synaptogenesis was the elimin-
ation of multiple synapses on spines and the ultimate
incorporation of postsynaptic membranes into en-
larged sites. The principle underlying compensatory
enlargement through elimination, may be dictated by
the postsynaptic neuron in its effort to maintain a
constant level of synaptic contact area (Hillman and
Chen, 1981).
6.2.2. Visual deprivation
Light deprivation causes changes in the spine num-
bers in the visual cortex of mouse (Valverde, 1967),
rabbit (Globus and Scheibel, 1967d) and rat (Fifkov~i,
1970) whereas enriched conditions produce increased
spine densities. Continuous exposure to light for 35
days produced increased spine densities on all den-
dritic segments but particularly the apical dendrites in
the visual cortex of rats (Parnavelas et al., 1973). If
each spine is the site of a synapse there will be
increased intracorticai connections and, thus, the
hyperstimulation-induced spine density increase may
reflect a greater demand for processing of infor-
mation. Valverde (1967) remarked that mice reared in
darkness (22-25 days) had structural changes in the
visual cortex including decreased apical spine den-
sities which were as a result of experience deprivation,
the result being a reduction in the effective contacts
of the dendritic spines. Although these values were
lower than controls the same exponential relationship
between the number of spines per segment and the
distance from the cell body was maintained. Valverde
(1967) suggested that the diminution of spines was
primarily a consequence of the arrested function of
the specific afferent fibres. The mechanism of spine
loss in this case was not thought to be related to
processes inducing transneuronal changes after le-
sioning such as deafferentation (Matthews and Pow-
ell, 1962), since the normal function of specific
afferents must be necessary for the maintenance of
the dendritic spine complex.