Content uploaded by David S Greenberg
Author content
All content in this area was uploaded by David S Greenberg on Jun 24, 2014
Content may be subject to copyright.
Imaging input and output of neocortical
networks
in vivo
Jason N. D. Kerr*, David Greenberg, and Fritjof Helmchen*
†
Department of Cell Physiology, Max Planck Institute for Medical Research, Jahnstrasse 29, D-69120 Heidelberg, Germany
Communicated by Bert Sakmann, Max Planck Institute for Medical Research, Heidelberg, Germany, July 19, 2005 (received for review June 1, 2005)
Neural activity manifests itself as complex spatiotemporal activa-
tion patterns in cell populations. Even for local neural circuits, a
comprehensive description of network activity has been impossi-
ble so far. Here we demonstrate that two-photon calcium imaging
of bulk-labeled tissue permits dissection of local input and output
activities in rat neocortex in vivo. Besides astroglial and neuronal
calcium transients, we found spontaneous calcium signals in the
neuropil that were tightly correlated to the electrocorticogram.
This optical encephalogram (OEG) is shown to represent bulk
calcium signals in axonal structures, thus providing a measure of
local input activity. Simultaneously, output activity in local neuro-
nal populations could be derived from action potential-evoked
calcium transients with single-spike resolution. By using these OEG
and spike activity measures, we characterized spontaneous activity
during cortical Up states. We found that (i) spiking activity is sparse
(<0.1 Hz); (ii) on average, only ⬇10% of neurons are active during
each Up state; (iii) this active subpopulation constantly changes
with time; and (iv) spiking activity across the population is evenly
distributed throughout the Up-state duration. Furthermore, the
number of active neurons directly depended on the amplitude of
the OEG, thus optically revealing an input–output function for the
local network. We conclude that spontaneous activity in the
neocortex is sparse and heterogeneously distributed in space and
time across the neuronal population. The dissection of the various
signal components in bulk-loaded tissue as demonstrated here will
enable further studies of signal flow through cortical networks.
bulk loading 兩 population imaging 兩 presynaptic 兩 sparse coding
U
nderstanding how information is represented and processed in
the mammalian neocortex requires measurement not only of
single-cell dynamics but also of spatiotemporal activity patterns in
identified networks of neurons in vivo. So far, optical imaging of
intrinsic or voltage-sensitive dye signals has revealed spatiotempo-
ral dynamics on the scale of cortical columns but has lacked cellular
resolution (1). Extracellular recording methods have enabled si-
multaneous measurements from multiple cells but suffer from
poorly defined cell identities, lack of spatial resolution, and are
incapable of resolving nonactive neurons (2). These techniques thus
fall short on providing a comprehensive description of cortical
microcircuits. In particular, these methods cannot monitor the
activation of afferent axons that represent the input into a particular
local region. Of key importance to a further understanding of
information processing in the neocortex will be a method that is
capable of simultaneously resolving both input and output of
cortical microcircuits with single-cell and single-spike resolution.
Here we apply recently developed techniques for two-photon
calcium imaging of neocortical cell populations in vivo (3–5) to
characterize neocortical activity during Up- and Down-state fluc-
tuations, which are observed spontaneously during anesthesia (6, 7),
sleep, and quiet wakefulness (8). We demonstrate that calcium
imaging of neuronal somata reveals local spike patterns with
single-cell and single-spike resolution. In addition, a prominent
axonal-based calcium signal was found in the neuropil, providing a
measure of local input. Analysis of the spatiotemporal distribution
of spiking activity during spontaneous Up states revealed that
spiking is sparse and heterogeneously distributed across neuronal
populations. Furthermore, we show that the number of spikes
depends on the level of driving input, as measured from the bulk
neuropil calcium signal. Thus, calcium imaging of bulk-labeled
tissue permits optical measurement of input–output relationships
in local cortical circuits in vivo.
Methods
Surgical Procedures. All experimental procedures were carried out
ac cording to the animal welfare guidelines of the Max Planck
Societ y. Wist ar rats (n ⫽ 35; P25–36) were i.p. anesthetized with
2 g of urethane per kg of body weight. The animal skull was
ex posed and cleaned, and a metal plate was attached to the skull
with dental acrylic cement. A 2- to 3-mm-wide craniotomy was
opened above either the primary motor cortex or somatosensory
c ortex. The exposed cortex was superfused with war m normal rat
ringer solution (135 mM NaCl兾5.4 mM KCl兾5 mM Hepes兾1.8
mM CaCl
2
, pH 7.2, w ith NaOH). The craniotomy was filled with
agarose (type III-A, Sigma; 1% in normal rat ringer solution)
and covered with an immobilized glass coverslip.
Labeling Procedures. Multicell bolus loading of neocortical cells with
the calcium indicator Oregon green 488 1,2-bis(2-aminophe-
noxy)ethane-N,N,N⬘,N⬘-tetraacetate-1 (OGB-1) acetoxymethyl
(OGB-1-AM; Molecular Probes) was performed as described in
refs. 3–5. In most experiments, multicell bolus loading was per-
formed in superficial layer 2兾3 (L2兾3). For specific loading of
dendritic structures in layer 1 and L2兾3, OGB-1-AM (1.4 mM) was
slowly pressure-ejected into layer 5 of motor cortex for 10–15 min
at 0.2–0.3 bar (1 bar ⫽ 100 kPa) with a micropipette tip 650–750
m below the pia. This deep loading re sulted in discrete labeling of
layer 5 pyramidal neurons. Dendrites could be imaged 1–1.5 h after
the injection. In vivo labeling of astrocytes was performed as
described in ref. 4.
Two-Photon Microscopy. Two-photon imaging was performed by
using a custom-built two-photon laser-scanning microscope as
described in refs. 4 and 9. Excitation wavelength was ⬇880 nm
(Mira 900-F laser, Verdi-10 pump, Coherent, Santa Clara, CA).
A n Oly mpus (Melville, NY) 20⫻ water-immersion objective lens
(0.95 numerical aperture) was used.
Electrophysiology. Electroc orticogram (ECoG) was recorded
with the tip of a 500-
m-diameter, Teflon-coated silver wire
placed against the pial surface in one corner of the cran iotomy.
A reference electrode was placed over cerebellum through a
small hole in the occipital bone. ECoG signals were acquired
Freely available online through the PNAS open access option.
Abbreviations: AMPA,
␣
-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid; AP, action
potential; ECoG, electrocorticogram; GYKI53655, 1-(4-aminophenyl)-3-methylcarbamyl-4-
methyl-3,4,-dihydro-7,8-methylenedioxy-5H-2,3-benzodiazepine; IEI, interevent interval;
OEG, optical encephalogram; OGB-1, Oregon green 488 1,2-bis(2-aminophenoxy)ethane-
N,N,N⬘,N⬘-tetraacetate-1; OGB-1-AM, OGB-1-acetoxymethyl; L2兾3, layer 2兾3.
*Present address: Department of Neurophysiology, Brain Research Institute, University of
Zurich, Winterthurerstrasse 190, CH-8057 Zu¨ rich, Switzerland.
†
To whom correspondence should be addressed. E-mail: helmchen@hifo.unizh.ch.
© 2005 by The National Academy of Sciences of the USA
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0506029102 PNAS
兩
September 27, 2005
兩
vol. 102
兩
no. 39
兩
14063–14068
NEUROSCIENCE
with a custom-built AC-coupled amplifier (input impedance 1
M⍀; bandwidth 0.1 Hz to 8 kHz).
Patch-clamp recordings were perfor med from OGB-1-AM
loaded L2兾3 neurons in either cell-attached or whole-cell con-
figuration. Cells were visually targeted by using a two-photon
microsc ope. Open pipette resistance was 4–6 M⍀. Intracellular
signals were recorded with an Axoclamp 2-B amplifier (Axon
Instr uments, Foster City, CA) and digitized with a CED1401plus
(Cambridge Electronic Design, Cambridge, U.K.).
Pharmacology. To locally block postsynaptic activity, we applied the
specific
␣
-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid
(AMPA)-type glutamate receptor antagonist 1-(4-aminophenyl)-
3-methylcarbamyl-4-methyl-3,4,-dihydro-7,8-methylenedioxy-5H-
2,3-benzodiazepine (GYKI53655) at 1 mM in normal rat ringer
solution. A micropipette containing GYKI53655 was inserted in
cortical L2兾3. After a control imaging period of 3 min, GYKI53655
was continually pressure-ejected (0.3–0.4 bar) for 3 min. The
application was ceased by either applying slight negative pressure
(⫺0.01 bar) or by withdrawing the pipette. Imaging continued for
a further 3–4 min.
Data Analysis. Calcium transients were measured by using f rames
made up of 64 ⫻ 128 or 32 ⫻ 128 pixels, with a line scan duration
of 1–1.5 ms (15- to 31-Hz frame rate). OGB-1 fluorescence was
averaged in regions of interest, including cell bodies or large
areas devoid of cell bodies (neuropil). Backg round fluorescence
was measured in unst ained blood vessels. Signals were ex pressed
as relative fluorescence changes (⌬F兾F) after background sub-
traction. Action potential (AP)-evoked calcium transients were
detected with a template-matching algorithm that took advan-
t age of their characteristic shape with a sharp rise followed by an
ex ponential decay (10). See the supporting information, which is
published on the PNAS web site, for more det ails on methods.
A ll data are presented as mean ⫾ SEM.
Results
Spontaneous Calcium Signals in L2兾3 of Neocortex. We investigated
spont aneous activit y in the neocortex by using the recently
developed multicell bolus loading technique for calcium indica-
tor loading of cell populations in vivo (3). The membrane-
per meable calcium indicator OGB-1-AM was pressure-ejected
through a glass pipette into L2兾3 of either motor cortex or barrel
c ortex in anaesthetized rats. Typically1hafterdyeinjection, all
cells within a radius of several hundred microns were labeled
(Fig. 1A ). Notably, not only cell bodies were stained but also the
neuropil, which c onsists of dendrites, axons, presynaptic bou-
tons, and glial processes. This staining pattern indicates that
various cellular compartments had taken up calcium indicator
dye. Neuropil staining was diffuse with no discernible structures,
presumably because of the lack of contrast between similarly
loaded subcellular compartments.
Unspecific loading by using the multicell bolus loading technique
necessitates a dissection of the various calcium signal components.
We previously showed that astrocytes can be identified in vivo by
using the red fluorescent dye sulforhodamine 101 (4). This marker
enables counterstaining of the astrocytic network and discrimina-
tion between the distinct calcium dynamics in astrocytes and
neurons, respectively (Fig. 1). As in our previous study we observed
slow calcium oscillations on the minute time scale in identified
astrocytes (mean half duration, 75 ⫾ 23 s; n ⫽ 15 cells and 5
animals). In contrast, neurons displayed spontaneous but infre-
quent calcium transients of short (⬍1 s) duration with fast onsets
and exponential decays, resembling AP-evoked calcium transients
observed in vitro and in vivo (9, 10). Most notably, large fluores-
cence changes were also present in the surrounding neuropil (Fig.
1B). Plotting the time course of mean fluore scence intensity in large
regions of interest not containing cell somata revealed intensity
fluctuations of 10–30% amplitude at frequencies in the 0.5–1 Hz
range (0.56 ⫾ 0.12 Hz, n ⫽ 5 animals). These fluctuations were
above the intrinsic noise level of our imaging system, suggesting that
they might represent a bulk measurement of calcium signals in
neuropil structures. Before further characterizing spontaneous
activity, we therefore addressed two crucial issues: (i) whether
single APs can be detected in individual neurons and (ii) what the
origin of the neuropil fluctuations is.
Single APs Are Resolved in Bulk-Loaded Tissue. To determine the
sensitivity of our measurements of neuronal calcium transients we
performed simultaneous cell-attached recordings (Fig. 2). Individ-
ual L2兾3 neurons were targeted with a patch pipette filled with the
red fluore scent dye Alexa Fluor 594, and a Gigaohm seal was
formed (Fig. 2A). Subsequently, spontaneous APs were recorded
extracellularly while somatic calcium transients in the same cell
were simultaneously measured (Fig. 2B). At the end of the record-
ing, a whole-cell configuration was e stablished to verify which
neuron had been recorded. To quantify the reliability of spike
detection, we analyzed the fraction of optically detected individual
APs and short bursts of a few APs (Fig. 2C). Overall, 97% of single
APs and 100% of bursts were detected (95 single APs, 46 doublets,
6 triplets, and 4 quadruplets; n ⫽ 14 cells and 12 animals) (Fig. 2E).
The calcium-transient amplitude correlated with the number of
APs (R
2
⫽ 0.81; see Fig. 4D) (11, 12). The average amplitude of
single-AP evoked transients was 10.0 ⫾ 0.9% and the mean decay
Fig. 1. Spontaneous calcium transients in cell somata and neuropil of
bulk-loaded neocortical L2兾3. (A)(Left) Side projection of OGB-1-loaded cells
in the motor cortex. Astrocytes (yellow) were counterstained with sulforho-
damine 101. (A)(Right)(Upper) Two-photon image 250
m below pial surface
showing neurons (green), astrocytes (yellow), and surrounding neuropil
loaded with OGB-1. (Lower) Same area as Upper showing regions of interest:
neuron (white circle), astrocyte (white square), neuropil (shaded gray), and
blood vessel lumen for background (blue circle). (B) Simultaneous calcium
transients from identified astrocyte (Top), neuron (Middle), and neuropil
(Bottom) recorded over several minutes. (Inset) Note the sharp transients with
fast onset and exponential decay (black) and ongoing neuropil signal on
expanded time scale (from boxes).
14064
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0506029102 Kerr et al.
time constant was 0.82 ⫾ 0.42 s (n ⫽ 10 transients and 5 animals).
Bursts of two or more APs resulted in larger calcium transients
(mean amplitude for doublets 15 ⫾ 2%; decay time constant 1.01 ⫾
0.3 s; n ⫽ 10 transients and 5 animals). We conclude that AP activity
is reliably resolved with single-cell and single-AP resolution by using
calcium imaging of bulk-loaded tissue. This makes it possible to
optically extract AP patterns, representing ‘‘output’’ activity, in local
neuronal circuits, providing an optical analogue to multiunit re-
cordings but with the advantage that AP activity can be assigned to
identified neurons.
Optical Encephalogram (OEG) Recording from Neuropil. To explore
the origin of neuropil fluorescence fluctuations we combined in
vivo two-photon imaging with ECoG and whole-cell recordings.
In whole-cell recordings f rom individual cells in bulk-loaded
tissue the membrane potential fluctuated between Up and Down
st ates as typically observed in neocortical neurons (Fig. 3A)
(6–8). This ongoing spontaneous activit y was also apparent in
the ECoG, which correlated with the intracellular recording
(peak correlation, 0.79 ⫾ 0.1; n ⫽ 5 animals) (7, 13). Neuropil
fluorescence fluctuations were highly synchronous to this elec-
trical activity indicating that they reflect ongoing Up and Down
st ates (Fig. 3 A and B and Fig. 7). Cross-correlations revealed a
tight c orrelation with both the ECoG and intracellular record-
ings (mean peak correlation of 0.70 ⫾ 0.1 and 0.75 ⫾ 0.1,
respectively; n ⫽ 5 an imals). For calcium indicator loading in
L2 兾3, both the peak ⌬F兾F amplitude of neuropil fluctuations
and the peak correlation with the ECoG did not depend on
depth f rom pial surface (Fig. 3 B, D, and E). These findings
c orroborate the idea that fluctuations in the neuropil are caused
by spontaneous calcium signals in its components and directly
linked to ongoing electrical activity. Given the strong correlation
with the ECoG we have termed the neuropil signal OEG.
OEG Originates from Axonal Structures. Neuropil in neocortex
c onsists of ⬎50% axons and presynaptic boutons, 30% dendrites
and dendritic spines, and 10% glial processes (14). All these
Fig. 2. Single AP-evoked calcium transients are detected in bulk-loaded tissue.
(A) Overlay of OGB-1-loaded neurons (green) and astrocytes (yellow). The cell-
attached recording was obtained from the center neuron (arrow indicates pi-
pette containing 5
M Alexa Fluor 594). (B)(Upper) Current trace from cell-
attached recording showing spontaneous spikes from neuron depicted in A
(number of spikes are indicated). (Lower) Simultaneously recorded somatic cal-
cium transients corresponding to single APs as well as AP doublet. (C) Examples
of somatic calcium transients (lower traces) evoked from spontaneous single APs,
double APs, and triple APs from cell-attached recordings (upper traces). (D)
Fidelity of spike detection confirmed with simultaneous cell-attached current
recordings. The fraction of optically detected single APs and bursts of APs is
shown. (E) Calcium transient amplitude as a function of the number of APs as
detected from simultaneous cell-attached current recordings.
Fig. 3. Neuropil fluctuations are highly correlated with ongoing electrical
activity. (A) Simultaneous measurement of membrane potential from a L2兾3
pyramidal neuron using whole-cell (WC) recording, ECoG, and neuropil fluo-
rescence fluctuation (OEG). (B)(Left) Side projection of cells superficially
loaded with OGB-1 depicting depth from pia that neuropil signals were
collected (dotted white lines). (Right) Ongoing OEG fluctuations (red) from
different depths correlated with electrical ECoG signals (black). (C) Dendrites
contribute little to OEG fluctuations. (Left) Side projection of OGB-1-loaded
dendrites originating from layer 5 neurons using deep-loading technique (see
Methods). (Right) Simultaneous ECoG recording (black) and OEG recording
from dendrites (red). The lower traces show a recording near to the site of
initial dye ejection (imaging depth indicated). (D) Peak amplitudes of neuropil
calcium transients in superficially loaded cortex (
■
) and deeply loaded cortex
(
E
). (E) Plot of peak correlation of neuropil and ECoG fluctuations at different
depths when OGB-1 is loaded into L2兾3(
■
, n ⫽ 5 animals) or discretely loaded
into layer 5 (
E
, n ⫽ 5 animals).
Kerr et al. PNAS
兩
September 27, 2005
兩
vol. 102
兩
no. 39
兩
14065
NEUROSCIENCE
str uctures potentially could c ontribute to OEG fluctuations.
Astroc ytes are unlikely to contribute because they display cal-
cium oscillations on a much slower time scale (Fig. 1) (4).
Moreover, the tight correlation between electrical activity and
OEG points to a neuronal origin. The question arises whether
the OEG represents mainly axonal (presynaptic) or dendritic
(postsynaptic) calcium signals.
To address this question, we used two complementary ap-
proaches. First, we developed a variant of the bulk-loading tech-
nique in which dye was discretely pre ssure-ejected in layer 5,
predominantly labeling deep cells and their apical dendrites that
project to upper cortical layers (Fig. 3C; see Methods). This loading
method enabled us to investigate the dendritic component of the
OEG in isolation. By using deep loading, the OEG was absent or
markedly reduced in superficial layers (Fig. 3C). Peak ⌬F兾F
amplitudes were 1.7 ⫾ 1.1% at a depth of 150
m and 3.7 ⫾ 2.9%
at 350
m, which was significantly lower than with superficial
loading (P ⬍ 0.001 for 150- to 450-
m depth; P ⬍ 0.01 for 550-
m
depth; n ⫽ 5 animals; paired t test) (Fig. 3D). Only at depths greater
than 500
m and near the injection site were OEG amplitudes
comparable with superficial loading (16.3 ⫾ 4.0%; P ⬎ 0.05 for 600-
to 680-
m depth). Similarly, by using deep loading, the correlation
peak between simultaneously recorded OEG and ECoG was
significantly reduced in the superficial layers down to 400
m but
not further below (peak correlation 0.1–0.2 at 50- to 150-
m depth,
0.2–0.4 at 200- to 400-
m depth, and 0.6–0.75 at ⬇600-
m depth)
(Fig. 3E). These results demonstrate that dendritic calcium tran-
sients contribute little if at all to the neuropil fluore scence signal,
suggesting that the OEG is mainly axonal in origin.
To further test this axonal origin hypothesis, we locally blocked
postsynaptic activity with GYKI53655, a specific antagonist of
AMPA-t ype glut amate receptors (15), which should min imally
af fect axonal calcium signals. Pressure ejection of 1 mM
GYKI53655 through a micropipette into L2 兾3 did not change the
amplitude of ongoing OEG fluctuations in this area (P ⫽ 0.87;
n ⫽ 4 animals) (Fig. 4 A and B). As a control, the postsynaptic
block ing effect of the antagonist was obvious in the simulta-
neously measured pattern of spont aneous AP-evoked calcium
transients (Fig. 4A). During antagonist application, the mean
f requency of calcium transients significantly decreased (0.038 ⫾
0.005 Hz before and 0.014 ⫾ 0.003 Hz with GYKI53655; P ⬍
0.001, paired test; n ⫽ 4 animals) (Fig. 4C) but recovered to
preant agonist levels af ter ter mination of drug application
(0.046 ⫾ 0.002 Hz; P ⬎ 0.05).
Taken together, these findings show that OEG fluctuations
predominantly originate from axonal structures, presumably
reflecting the bulk average of AP-evoked calcium transients in
presynaptic boutons and axons activated during the Up-state
periods. Thus, the OEG can be considered a measure for
volume-averaged ‘‘input’’ activity to the particular local region.
APs Are Uniformly Spread Throughout Cortical Up States. In vivo
calcium imaging in bulk-loaded tissue thus permits measurement
of input and output neuronal activit y in local cortical areas. We
used this approach to characterize ongoing spontaneous activity
in the neoc ortex on a population level. We first investigated the
temporal str ucture of AP activ ity during Up states. In whole-cell
rec ordings f rom L2 兾 3 pyramidal neurons, APs occurred with a
un iformly distributed temporal delay with respect to the onset of
Up states (Fig. 5A). To investigate whether such a uniform
distribution holds true at the population level, we perfor med a
similar analysis on the somatic calcium transients detected
simult aneously in small neuronal populations (up to 40 neurons).
Fig. 5B shows 20 consecutively recorded Up states, as resolved
f rom the ECoG and aligned to their onset, and corresponding
AP-evoked calcium transient patterns of 38 identified neurons.
Note that not every Up state produced AP-evoked calcium
transients within the recorded neuronal population. As in the
whole-cell recordings, AP activity was spread over the entire Up
st ate and was not confined to one particular period (Fig. 5B).
Pooling the delays f rom onset of AP-evoked calcium transients
for 220 Up st ates from five an imals and temporally binning these
results revealed no significant difference between the different
time bins during the average Up-state time course (P ⬎ 0.05)
(Fig. 5C). Cumulative temporal distributions of both electrically
measured and optically detected AP activity were linear (R
2
⬎
0.99), indicating a uniform distribution.
Heterogeneity and Sparseness of Population Activity. We next aimed
to deter mine the spatiotemporal distribution of the active sub-
population of neurons that participate in Up-states. We per-
for med measurements in 90-s imaging periods from areas con-
t aining 20–30 neurons (Fig. 6A ). To reveal the spatial pattern of
activity we color-coded neurons according to the fraction of total
Up-st ates during which they evoked a calcium transient. We
found that the spatial organization of active neurons was not
st able but displayed considerable heterogeneity over the time
c ourse of minutes (Fig. 6B). This heterogeneity indicates that
spont aneous activity does not emerge exclusively in a particular
subset of neurons but rather is generated by a continually
changing subpopulation of neurons. Neurons also displayed
heterogeneit y with respect to their temporal activation. To test
this we analyzed the distribution and coefficient of variance of
interevent intervals (IEI) from individual neurons within a
net work (Fig. 6E). An event was defined as a calcium transient
generated by one or more APs. The IEI distribution from
individual neurons over 6 min (data not shown) as well as IEI
distributions pooled from neuronal populations could be fitted
by an ex ponential curve (
⫽ 15.7 s, R
2
⫽ 0.98) (Fig. 6E). IEI
f rom individual neurons displayed high c oefficient of variance
values that ranged from 0.58 to 1.5 w ith a mean of 0.95 ⫾ 0.04
Fig. 4. Blocking postsynaptic spiking activity does not change OEG. (A) Spike-
evoked calcium transients (raster plot) from 25 L2兾3 neurons before, during (gray
box), and after local pressure application of a specific AMPA receptor antagonist
(1 mM GYKI53655). Shownare group activity during thesame time periods (lower
histogram; 1-s bins) and total activity of individual neurons (upper right histo-
gram). Neuropil calcium fluctuations (red) recorded during the same period as
raster plot. (B) Representative ECoG (black trace) and neuropil calcium fluctua-
tion (red trace) periods and resulting cross correlations before and during antag-
onist application (taken from dotted boxes). (C) Summary of neuronal firing rates
(black) and peak correlations (red) comparing preantagonist periods (pre) to
antagonist periods and postantagonist periods.
14066
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0506029102 Kerr et al.
(n ⫽ 5 animals). These results suggest that AP activit y during
spont aneous ongoing activit y in its simplest for m is Poisson-
distributed.
A major advantage of the imaging approach is that it per mits
the identification of nonactive cells, which enabled us to deter-
mine the f raction of cells that were active per individual Up state.
In individual experiments, this fraction varied between 0% and
50% (Fig. 6C). Pooling many Up states from several animals
yielded a skewed distribution with an average 10.6 ⫾ 2.1% of
neurons being active per Up state (183 neurons; 5 animals).
Neurons exhibited low rates of somatic calcium transients of
between 0 and 0.14 Hz with a mean value of 0.048 ⫾ 0.002 Hz (n ⫽
212 neurons and 11 animals) (Fig. 6F). These rates are consistent
with the low rates of spontaneous APs observed in electrical
recordings (cell-attached recordings: 0.05 ⫾ 0.008 Hz, n ⫽ 14
neurons and 12 animals; whole-cell recordings: 0.05 ⫾ 0.01 Hz, n ⫽
10 neurons and 9 animals; no significant difference between all
groups, P ⬎ 0.05). Although some spontaneous calcium transients
may have been evoked by bursts of several APs, these events were
rare (⬍10%) and therefore had a negligible effect on the calculated
rates using optical data. Of identified neurons, 11% did not show
any calcium transients during the imaging period. One possibility is
that the se cells are interneurons, for which it is not clear whether
single APs can be resolved. Ranking of all neurons according to
their mean activity revealed a continuous, smooth distribution
ranging from nonactive cells to cells with relatively high activity
(Fig. 6F). Together, these results show that spontaneous AP activity
in cortical L2兾3 neurons is sparsely and heterogeneously distributed
in space and time. The level of activity varie s throughout the
population but rather in the form of a continuum than in discrete
groups of nonactive and active neurons.
Optical Recording of Local Input–Output Relationship. The OEG and
the AP-evoked calcium transient patterns reflect input and output
activity, respectively, in a local area of neocortex. A larger input, i.e.,
a stronger axonal activation, therefore should result in a larger
output, i.e., generation of more APs. We te sted this hypothesis by
plotting the number of postsynaptic calcium transients in local
groups of neurons against the amplitude of the OEG in the
surrounding neuropil as a measure of the strength of axonal
activation to this particular region. This analysis revealed a depen-
dence resembling that of an input–output relationship (16, 17) with
a threshold below which no APs were generated and an approxi-
mately linear relationship above threshold (n ⫽ 5 animals) (Fig.
6H). The ability to resolve axonal and cellular calcium signals in
bulk-loaded tissue thus promises to be a useful tool for studying
signal transformation in local neocortical networks.
Fig. 5. Distribution of activity during Up states. (A) Overlay of three spon-
taneous whole-cell Up states (Upper) and the simultaneous ECoG recordings
(Lower) depicting variable timing of AP firing with respect to Up-state onset.
Shown is the distribution of AP firing times from 43 Up states (n ⫽ 5 animals)
with reference to Up-state onset (Lower). Indicated are the first (
■
), second
(
䊐
), and third (
E
) APs. APs from upper membrane potential traces are indi-
cated by colored symbols. (B) Raster plot of somatic calcium transients in 38
neurons imaged during 20 consecutive spontaneous ECoG Up states (overlay).
Each line represents peak time of the activity, and the color is associated with
the color of the different Up state (overlay). (C) Histogram of activity from 220
Up states from five animals showing average (red line) time of postsynaptic
activity in relation to the average Up state time course (0.1-s bins).
Fig. 6. Heterogeneous population spiking activity. (A) Fluorescence image
showing the spatial distribution of the astrocytes (yellow) and OGB-1-filled
neurons (green). (B) Pseudocolored representation of imaged area in A de-
picting the fraction of Up states in which neurons were active during a 90-s
period. Activity scale is shown; neurons that were not active within this period
are colored black. (C) Raster plot depicting the percentage of neurons from
one area that produce spike-evoked calcium transients during consecutive Up
states. Each point represents one Up state. The average activity is depicted by
the red line. (D) Population histogram for the average percentage of neurons
active during many Up states from five animals. (E) Population IEI distribution
pooled from many imaging periods (exponential fit, red). (F) Range of firing
frequencies in which 212 neurons were active over a period of 10 min; neurons
were ranked according to their average activity (red point indicates mean). (G)
Probability of spiking activity in a population of neurons from five animals
(same data as in E). (H) Local input– output relationship. Relative ⌬F兾F changes
in OEG fluorescence (presynaptic) during an Up-state transition calculated
from the preceding Down state and plotted against the absolute number of
postsynaptic spiking-related events (postsynaptic) during the corresponding
Up state over the entire neuronal population.
Kerr et al. PNAS
兩
September 27, 2005
兩
vol. 102
兩
no. 39
兩
14067
NEUROSCIENCE
Discussion
Optical Measurement of Output Activity. APs cause calcium influx
through voltage-dependent calcium channels and evoke changes in
intracellular calcium concentration. Hence, calcium imaging from
cell populations can be used to extract spiking (output) patterns in
neuronal networks, as has previously been shown in brain slices
(18). Here, we extended this approach to in vivo conditions and
demonstrated that single spikes can be optically detected after bulk
loading with OGB-1. Such ‘‘optical multiunit recording’’ with
single-cell resolution permits the extraction of spiking patterns in
groups of up to 40 neurons at an ⬇10-Hz frame rate. The temporal
precision of this approach currently is limited because of the frame
acquisition rate and the relatively slow (1 s) decay of the calcium
signal, which causes calcium transients to merge as APs occur in
rapid succession (10, 18). Taking into account the amplitude of
AP-evoked calcium transients may improve the temporal accuracy
of spike pattern extraction, because the amplitude depends on the
number of APs (19) (Fig. 2E), their relative timing (20), and their
spacing relative to the Up-state onset (21). Alternatively, faster
dyes, such as voltage-sensitive dyes, could increase temporal re so-
lution (1), but these dyes currently lack cellular resolution in vivo.
In the current study, only a small fraction of events consisted of
more than one AP, suggesting that the limited temporal resolution
did not adversely affect the conclusions drawn.
Optical Measurement of Input Activity. A major finding of our study
is the prominent spontaneous calcium signal in the neuropil,
which we have termed OEG. We demonstrated that the OEG is
predominately axonal in nature, c onsistent with the large frac-
tion of neuropil volume occupied by axons and presynaptic
boutons (14, 22). In our view, the OEG represents a volume-
averaged signal, reflecting the summation of AP-evoked calcium
transients in many activated axons. The OEG is thus comple-
ment ary to the ECoG, which is thought to reflect synaptic
potentials. Given that compound synaptic potentials reflect the
mean rates of presynaptic input, the high correlation between
OEG and ECoG is not surprising.
Two lines of evidence supported a presynaptic axonal origin of
the OEG. First, dendrites in the neuropil contributed minimally to
the OEG, which is in agreement with single-cell imaging experi-
ments that reported no detectable calcium transients in apical
dendritic trunks during subthreshold Up states (23). Although
localized synaptic calcium transients in dendritic spines could
contribute, they largely depend on NMDA receptor activation and
therefore may be relatively small near resting membrane potential.
Second, the OEG was not affected by locally blocking AMPA
receptors, whereas the frequency of neuronal spike-induced cal-
cium transients were clearly reduced. Although L2兾3 pyramidal
neuronal axon collaterals are located in close proximity to the
neuron of origin and therefore may contribute to the OEG, no
significant changes in the OEG were detected when spiking of local
neurons was markedly reduced. In addition, neuronal firing was
infrequent, but the OEG was continuous, and if the OEG did, in
fact, rely on activity from local neuron axon arbors, then one would
expect a more discontinuous OEG signal and less correlation with
simultaneously recorded ECoG signals. Thus, OEG recording from
the neuropil allows monitoring of volume-averaged input into the
local neuronal network and correlation with simultaneous neuronal
output activity.
Characteristics of Spontaneous Cortical Activity. Both electrical and
optical recordings consistently revealed that individual neurons
as well as populations of neurons display sparse spont aneous
activity. Single neurons displayed low AP rates of ⬍0.1 Hz, in
agreement with previous in vivo studies (24, 25). On a population
level, only a fraction (10%) of neurons was active during any
given Up state, with this subset of active neurons changing over
time. This temporal and spatial sparseness suggests a further
capacit y of the neocortex for encoding additional signals, e.g.,
sensory-evoked inputs, during Up-state periods. IEI distribu-
tions from neuronal populations were well approximated by a
Poisson process. This stochastic behavior suggests that spikes do
not explicitly depend on previous spiking times during sponta-
neous activit y, which is in contrast to other studies that have
reported spatial and temporal spiking structure in acute cortical
slices (26, 27). Although these studies recorded transients from
many more cells than in the present study, neurons were not
distinguished from astrocytes, which c ould be import ant con-
sidering that astrocytes are gap–junction c oupled and display
calcium transients (4).
In summary, we have demonstrated that input and output activity
can be measured in a local cortical region by using bulk calcium
indicator loading. Furthermore, we observed a direct relationship
between input and output because the amount of postsynaptic
spiking activity depended on the strength of presynaptic axonal
activity, measured by the OEG. The investigation of input–output
relationships is of key importance for understanding signal pro-
cessing in neuronal circuits (16, 28). Our optical approach, for
example, could be used to investigate modulation of the gain of this
transformation under various conditions of excitatory and inhibi-
tory synaptic input (16). Furthermore, it should enable the study of
input–output transformations during sensory input and how they
are affected by varying levels of background activity such as they
occur during different behavioral states.
We thank B. Sakmann and W. Denk for generous support; Marlies
Kaiser for expert technical support; John D. Rolston for programming
initial analysis sof tware; and Drs. Tansu Celikel, Andreas Frick, and
Rainer Friedrich for comments on an earlier version of the manuscript.
We also thank Patrick Theer and Werner Go¨bel for useful discussions
and help with microscope modifications.
1. Grinvald, A. & Hildesheim, R. (2004) Nat. Rev. Neurosci. 5, 874–885.
2. Buzsaki, G. (2004) Nat . Neurosci. 7, 446–451.
3. Stosiek, C., Garaschuk, O., Holthoff, K. & Konnerth, A. (2003) Proc. Natl.
Acad. Sci. USA 100, 7319–7324.
4. Nimmerjahn, A., Kirchhoff, F., Kerr, J. N. & Helmchen, F. (2004) Nat. Methods
1, 31–37.
5. Ohki, K., Chung, S., Ch’ng, Y. H., Kara, P. & Reid, R. C. (2005) Nature 433,
597–603.
6. Cowan, R. L. & Wilson, C. J. (1994) J. Neurophysiol. 71, 17–32.
7. Lampl, I., Reichova, I. & Ferster, D. (1999) Neuron 22, 361–374.
8. Petersen, C. C., Hahn, T. T., Mehta, M., Grinvald, A. & Sakmann, B. (2003)
Proc. Natl. Acad. Sci. USA 100, 13638–13643.
9. Waters, J., Larkum, M., Sakmann, B. & Helmchen, F. (2003) J. Neurosci. 23,
8558–8567.
10. Helmchen, F., Imoto, K. & Sakmann, B. (1996) Biophys. J. 70, 1069–1081.
11. Helmchen, F., Borst, J. G. & Sakmann, B. (1997) Biophys. J. 72, 1458–1471.
12. Svoboda, K., Denk, W., Kleinfeld, D. & Tank, D. W. (1997) Nature 385, 161–165.
13. Mahon, S., Deniau, J. M. & Charpier, S. (2001) Cereb. Cortex 11, 360–373.
14. Chklovskii, D. B., Schikorski, T. & Stevens, C. F. (2002) Neuron 34, 341–347.
15. Paternain, A. V., Morales, M. & Lerma, J. (1995) Neuron 14, 185–189.
16. Chance, F. S., Abbott, L. F. & Reyes, A. D. (2002) Neuron 35, 773–782.
17. Fellous, J. M., Rudolph, M., Destexhe, A. & Sejnowski, T. J. (2003) Neuro-
science 122, 811–829.
18. Smetters, D., Majewska, A. & Yuste, R. (1999) Methods 18, 215–221.
19. Kerr, J. N. & Plenz, D. (2002) J. Neurosci. 22, 1499–1512.
20. Callaway, J. C. & Ross, W. N. (1995) J. Neurophysiol. 74, 1395–1403.
21. Kerr, J. N. & Plenz, D. (2004) J. Neurosci. 24, 877–885.
22. Braitenberg, V. & Schu¨z, A. (1998) Cortex: Statistics and Geometry of Neuronal
Connectivity (Springer, Berlin).
23. Waters, J. & Helmchen, F. (2004) J. Neurosci. 24, 11127–11136.
24. Brecht, M. & Sakmann, B. (2002) J. Physiol. 538, 495–515.
25. Margrie, T. W., Brecht, M. & Sakmann, B. (2002) Pflugers. Arch. 444, 491–498.
26. Cossart, R., Aronov, D. & Yuste, R. (2003) Nature 423, 283–288.
27. Ikegaya, Y., Aaron, G., Cossart, R., Aronov, D., Lampl, I., Ferster, D. & Yuste,
R. (2004) Science 304, 559–564.
28. Shadlen, M. N. & Newsome, W. T. (1998) J. Neurosci. 18, 3870–3896.
14068
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0506029102 Kerr et al.