Notes

Information is stored in neural circuits through long-lasting
changes in synaptic strengths1,2. Most studies of information
storage have focused on mechanisms such as long-term potentiation and depression (LTP and LTD), in which synaptic strengths
change in a synapse-specific manner3,4. In contrast, little attention
has been paid to mechanisms that regulate the total synaptic
strength of a neuron. Here we describe a new form of synaptic
plasticity that increases or decreases the strength of all of a
neuron’s synaptic inputs as a function of activity. Chronic blockade
of cortical culture activity increased the amplitude of miniature
excitatory postsynaptic currents (mEPSCs) without changing
their kinetics. Conversely, blocking GABA (g-aminobutyric
acid)-mediated inhibition initially raised firing rates, but over a
48-hour period mESPC amplitudes decreased and firing rates
returned to close to control values. These changes were at least
partly due to postsynaptic alterations in the response to glutamate,
and apparently affected each synapse in proportion to its initial
strength. Such ‘synaptic scaling’ may help to ensure that firing
rates do not become saturated during developmental changes in
the number and strength of synaptic inputs5
, as well as stabilizing
synaptic strengths during Hebbian modification6,7 and facilitating
competition between synapses7–9
.
We tested the idea that postnatal rat visual cortical pyramidal
neurons in primary cell culture scale the strength of AMPA (aamino-3-hydroxy-5-methyl-4-isoxazole propionic acid)-mediated
synaptic currents up or down as a function of activity. To sample
inputs from a large number of synapses we used whole-cell voltageclamp recordings to measure mEPSCs. In control cultures there
was a broad distribution of mEPSC amplitudes10,11 (Fig. 1a;
average quantal amplitude ¼ 13:6 6 0:7 pA at −70 mV, n ¼ 27).
The effects of decreasing or increasing firing rates for 48 h on
mEPSC amplitude are shown in Fig. 1a. On average, mEPSC
amplitudes from cultures grown in tetrodotoxin (TTX), whichcompletely abolished firing, increased to 192 6 16% of control
values, and there was a similar effect on area (Fig. 1b, n ¼ 10).
Blockade of GABAA-mediated inhibition with bicuculline, which
raised firing rates to 265 6 10% of control values, significantly
decreased mEPSC amplitude to 70 6 4% of control values (Fig. 1b,
n ¼ 10), and also decreased mEPSC area. For the manipulations
described here and below, resting potentials, input resistances,
whole-cell capacitances and series resistances were similar between
conditions (Table 1). These data demonstrate that the quantal
amplitude of AMPA synapses can be increased or decreased by
changes in activity. Activity blockade did not significantly affect the
amplitude of AMPA-mediated mEPSCs recorded from bipolar
interneurons (Fig. 1c, n ¼ 14), indicating that excitatory inputs
from pyramidal neurons are regulated differently depending on the
nature of the target neuron.
Some forms of synaptic plasticity require calcium influx through
NMDA (N-methyl-D-aspartate) receptors3,4. To examine the effects
of NMDA receptor blockade on mEPSC amplitude, cultures were
grown in the NMDA antagonist AP5 (D(−)-amino-7-phosphonovalenic acid) for 48 h AP5 did not influence firing rates (95 6 24%
of control values), and had no effect on mEPSC amplitude or area
(Fig. 1b, n ¼ 8). In contrast, blocking AMPA receptors with 6-
cyano-7-nitroquinoxaline-2,3-dione (CNQX), which like TTX
completely abolished firing, significantly increased mEPSC amplitude to 195 6 24% of control values (Fig. 1b, n ¼ 10). These
observations indicate that, in contrast to many forms of LTP and
LTD3,4,12, the change in synaptic strength produced by activity
blockade is not simply the result of a change in NMDA receptor
signalling.
Cumulative amplitude histograms showed that the entire distribution of mEPSC amplitudes shifted towards larger values forTTX-treated and CNQX-treated cultures, and towards smaller
values for bicuculline-treated cultures (Fig. 1d). We examined
mEPSC kinetics by determining the average mEPSC waveforms
under the different conditions (Fig. 1e). Scaling and overlaying
these averaged waveforms revealed no differences in the rise and
decay kinetics (Fig. 1e), and individual measurements of rise times
also showed no significant differences between conditions (Table 1).
In contrast to LTP and other rapid potentiation protocols13–15, there
were no significant differences in mEPSC frequency between conditions (Table 1). The lack of effect on frequency and kinetics
suggests that, in accord with other studies16,17, altering activity for
two days does not significantly influence the placement or number
of excitatory synapses onto pyramidal neurons.
The time course of mEPSC regulation was determined by blocking activity for 15, 26, or 48 (64) h (n ¼ 8, 11 and 10, respectively).
There was a progressive increase in mEPSC amplitude with treatment time, indicating that this process is both slow and cumulative
(Fig. 2a).
Does the activity-dependent regulation of mEPSC amplitude
produce a change in spike-mediated EPSCs? TTX treatment increasedEPSC amplitude between pairs of monosynaptically connected
pyramidal neurons from 89 6 65 pA (measured at −70 mV,
n ¼ 13 pairs) to 219 6 52 pA (n ¼ 9 pairs) (Fig. 2b). There was
no effect on EPSC reversal potential (1:6 6 1:0 and 2:2 6 1:1 mV
for control and TTX-treated pyramidal neurons, respectively).
The regulation of mEPSC amplitudes described above could act
to stabilize firing rates, because raising firing rates decreases the
strength of excitatory synaptic inputs, and vice versa. To test
whether firing rates are regulated homeostatically in response to
changesin activity, cultureswere grown under control conditions or
in bicuculline for 48 h, and whole-cell current-clamp recordings
were used to measure firing rates (Fig. 2c). Acute bicuculline
treatment raised firing rates significantly from 0:4 6 0:1 to
1:1 6 0:2 Hz (Fig. 2d), but after 48 h in bicuculline, firing rates
recorded in bicuculline declined significantly to 0:24 6 0:06 Hz, and
were close to zero when bicuculline was removed (Fig. 2d). These
data indicate that increased firing produces a regulatory response
that returns firing rates to control levels. Previous studies have
shown that chronic TTX treatment dramatically increases firing
rates when the TTX is removed18,19. The bidirectional regulation of quantal amplitude is likely to contribute significantly to the homeostatic regulation of firing rates.
Changes in mEPSC amplitude could be occurring through a
postsynaptic change in glutamate responsiveness or a presynaptic
change in the glutamate content of synaptic vesicles. To investigate
this, pulses of glutamate were applied to pyramidal neurons in the
presence of TTX and AP5 and the amplitudes of the resulting
glutamate currents were measured (Fig. 3a, b). Chronic TTX
treatment increased the current amplitude to 278 6 39% of control
values at the soma, and similar increases were seen along the apical
dendrite (Fig. 3c; control, n ¼ 9; TTX, n ¼ 8). No difference in rise
times or times to peak of the glutamate currents were observed
between conditions, indicating that the difference in amplitude is
not due to a difference in rates of diffusion or removal of glutamate
under the two conditions (rise times ¼ 7:5 6 0:7 and 8:9 6 0:8 ms,
and times to peak ¼ 23 6 1:4 and 25 6 1:1 ms, respectively). These
data suggest that activity regulates mEPSC amplitude through a
postsynaptic change in receptor number orfunction, although there
may be additional presynaptic changes. In contrast to supersensitivity at the neuromuscular junction, where enhanced responsiveness is exclusively extrajunctional and mEPP amplitude does not
change20,21, the change in mEPSC amplitude observed here indicates
that receptors located at the synapse are involved in the regulatory
process.
What is the relationship between the mEPSC amplitude distributions produced by different activity levels? The average quantal
amplitude could be modified by changing synaptic strengths by a
constant value (an additive function), by adding a random amount
(a random additive function), or by scaling synaptic strengths by the
same multiplicative factor (a multiplicative function). To distinguish between these possibilities, control amplitudes were plotted
against TTX amplitudes (Fig. 4a) or bicuculline amplitudes (Fig.
4b) and the resulting relationship fit using each of the above models.
The data were well fit by linear functions with slopes of 2.73 (TTX)
and 0.66 (bicuculline), but not by additive or random additive
functions (Fig. 4a). When scaled multiplicatively by these slope
values, the cumulative amplitude distributions for TTX and bicuculline were almost perfectly superimposable on the control distribution (Fig. 4b), suggesting that activity is globally scaling
quantal amplitudes in a multiplicative manner. Taken together,
our data place several constraints on the biophysical mechanism of
synaptic scaling. If this scaling occurs through a change in the
function of existing AMPA receptors through phosphorylation22,23
or some other mechanism, this process must modify single-channel
conductance without influencing mEPSC kinetics. Alternatively, if
this scaling occurs through changes in the synthesis and insertion of
glutamate receptors24,25, or through conversion of existing receptors
between inactive and active states26,27, then receptors must be
inserted or converted in proportion to the existing number of
functional receptors.
Multiplicative scaling of synaptic strengths preserves relative
differences between inputs (such as those produced by Hebbian
modifications), while allowing a neuron to adjust the total amount
of synaptic excitation it receives. This process may help developing
neurons remain responsive to inputs both early in development,
when the number of synaptic inputs is small, and later in development as the number rises28. This may be important for allowing
developing neuronsto participate in the correlation-based mechanisms that contribute to circuit formation5
. Consistent with this,
there is evidence that mEPSC amplitudes in hippocampal neurons
decrease as the number of synaptic inputs increase29. In addition, by
contributing to a stabilization in firing rates, synaptic scaling may
help to counteract the destabilizing effects of Hebbian synaptic
modifications6,7,30. This instability arises because synaptic potentiation increases postsynaptic firing rates, thus increasing the correlation with presynaptic activity and producing a positive feedback
loop that eventually saturates evenweakly correlated inputs. Finally this process predicts that strengthening of some inputs will lead to a
scaling down of all synaptic strengths. Synaptic weakening has been
suggested to be the prelude to synapse elimination8,9. By reducing
weak inputs in response to the strengthening of others, synaptic
scaling may contribute to the processes of synaptic competition and
elimination.