report - university of michigansites.lsa.umich.edu/aton-lab/wp-content/uploads/... · report gaba...
Post on 16-Apr-2020
3 Views
Preview:
TRANSCRIPT
Neuron
Report
GABA Networks Destabilize Genetic Oscillationsin the Circadian PacemakerG. Mark Freeman, Jr.,1 Rebecca M. Krock,1 Sara J. Aton,1 Paul Thaben,2 and Erik D. Herzog1,*1Department of Biology, Washington University, St. Louis, MO 63130, USA2Institute for Theoretical Biology, Humboldt University, 10115 Berlin, Germany
*Correspondence: herzog@wustl.eduhttp://dx.doi.org/10.1016/j.neuron.2013.04.003
SUMMARY
Systems of coupled oscillators abound in nature.How they establish stable phase relationships underdiverse conditions is fundamentally important. Themammalian suprachiasmatic nucleus (SCN) is aself-sustained, synchronized network of circadianoscillators that coordinates daily rhythms in physi-ology and behavior. To elucidate the underlyingtopology and signaling mechanisms that modulatecircadian synchrony, we discriminated the firing ofhundreds of SCN neurons continuously over days.Using an analysis method to identify functionalinteractions between neurons based on changes intheir firing, we characterized a GABAergic networkcomprised of fast, excitatory, and inhibitory connec-tions that is both stable over days and changes instrength with time of day. By monitoring PERIOD2protein expression, we provide the first evidencethat these millisecond-level interactions activelyoppose circadian synchrony and inject jitter into dailyrhythms. These results provide a mechanism bywhich circadian oscillators can tune their phase rela-tionships under different environmental conditions.
INTRODUCTION
The study of coupled oscillators is part of a broader movement
toward understanding complex systems (Strogatz, 2000). In
biology, intercellular communication can modulate the precision
and synchronization of single-cell oscillations including glycol-
ysis, somitogenesis, respiration, and daily cycling (Jiang et al.,
2000; Herzog, 2007). In many cases, coupled systems are inher-
ently difficult to understand because the interactions are diverse
and dynamic. The suprachiasmatic nucleus (SCN) of the
mammalian brain provides an exceptional opportunity to reveal
the topology, types, stability, and function of diverse connec-
tions in a defined network of neural oscillators.
Neurons within the SCN express near 24 hr (circadian)
oscillations in electrical activity and gene expression and entrain
to regulate daily rhythms including metabolism, hormone
release, and sleep-wake cycles. These cells depend on an intra-
cellular transcription-translation feedback loop to generate daily
rhythms and intercellular signaling for both synchronization and
reliable rhythmicity (Yamaguchi et al., 2003; Webb et al., 2009).
Importantly, it is unknown how intercellular signaling modulates
the cycle-to-cycle precision of circadian rhythms.
Neural communication in the SCN includes gap junctions,
neurotransmitters and neuropeptides. Of these, loss of vasoac-
tive intestinal polypeptide (VIP) dramatically impairs circadian
rhythms in the SCN and in behavior (Aton et al., 2005). Recent
links between VIP signaling and schizophrenia highlight the
possibility that VIP determines the development of the circuits
underlying circadian synchrony (Vacic et al., 2011). To test
whether VIP is required to maintain network topology in the
SCN, we established a novel method to reliably map the func-
tional connections between SCN neurons.
Within the central nervous system, g-amino-butyric acid
(GABA) serves as the principal inhibitory neurotransmitter.
Nearly every neuron within the SCN synthesizes GABA (Moore
and Speh, 1993; Belenky et al., 2008) and exhibits inhibitory
postsynaptic currents (IPSCs) that depend on GABA signaling
and vary in frequency over the day (Itri et al., 2004). In spite of
its predominance, however, the function of GABAergic signaling
in the SCN remains unresolved. GABA has been reported to be
inhibitory at all times (Aton et al., 2006; Liu and Reppert, 2000),
mainly inhibitory during the day and excitatory during the night
(Albus et al., 2005; Choi et al., 2008; De Jeu and Pennartz,
2002) and inhibitory during the night, excitatory during the day
(Wagner et al., 1997). Furthermore, daily administration of exog-
enous GABA suffices to coordinate SCN neurons (Liu and
Reppert, 2000), and GABA can transmit phase information
between SCN populations (Albus et al., 2005); however, syn-
chrony among SCN cells can persist during chronic blockade
of intrinsic GABAergic signaling (Aton et al., 2006). To resolve
these apparent contradictions, we discriminated the discharge
patterns of large numbers of individual neurons over multiple
days and identified the stability and polarity of GABA-dependent
interactions in the SCN. Using real-time bioluminescence
imaging, we discovered a role for these synapses in circadian
timekeeping.
RESULTS
Spontaneous Firing Reveals Fast ConnectivityTo assess functional communication between SCN neurons, we
monitored gene expression and firing rates of individual SCN
neurons in vitro. We found that in explants and dispersals,
Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc. 799
Figure 1. Detection and Mapping of GABA-Dependent interactions in SCN Networks
(A) SCN neurons were recorded on a MEA. Colored shapes show the locations where four representative neurons were recorded. (B) The firing patterns of these
neurons differed in their time of peak firing over the day (left) and in their spike patterns (right). (C) Spike trains from one recorded hour (gray box in B) cross-
correlated positively, negatively, or not at all between each pair of neurons. These predictable changes in firing were illustrated as excitatory (red arrow) or
inhibitory (blue arrow), directed connections on amicrograph of the SCN culture. Dashed lines represent significance thresholds as determined by BSAC and the
magnitude of each Z-score is color coded (graded red and blue bars on the right). In these examples, the probability of neuron 2 firing increased by 233% after
neuron 1 fired and the probability of neuron 4 firing decreased by 17% after neuron 3 fired. (D) A map of interactions detected from a representative 24 hr
recording. The 103 firing neurons (nodes) made 542 directed connections (arrows). (E) The Z score of each interaction as a function of the latency to the peak of its
cross-correlogram. Note from the cumulative distribution of these excitatory (red line) and inhibitory (blue line) interactions that most peaked between 2.5 and
15 ms, consistent with monosynaptic communication. (F) Treatment of SCN cultures with GABAA receptor antagonists (200 mM bicuculline or 100 mM gabazine)
significantly decreased the number of excitatory and inhibitory connections. The remaining correlations were relatively weak and accounted for less than 7% of
the connections found in vehicle-treated cultures. (G) Degree distributions of three representative SCN networks were not linear on log-log plots and therefore
deviate from the power law predicted for scale-free networks. In these representative networks, the likelihood P(k) of finding a neuronwith k outgoing connections
decreased exponentially with increasing connections and the averaged degree distribution across networks was best fit by a first-order exponential decay
function (r2 = 0.95, red line). Scale bars, 200 mm.
Neuron
GABA Networks Destabilize the Circadian Pacemaker
SCN neurons maintained synchronized circadian rhythms for as
long as we recorded, demonstrating that the network mecha-
nisms underlying coordinated circadian rhythmicity are intrinsic
to these cultures (see Figure S1 available online).We took advan-
tage of this self-sustained neural circuit to test the role and sta-
bility of specific connections in circadian rhythms. We recorded
spontaneous action potentials frommany SCN neurons simulta-
neously and continuously over days with 40 ms resolution on
multi-electrode arrays (MEAs) (Figure 1A). Consistent with previ-
ous reports (Welsh et al., 1995), circadian neurons fired daily for
9.8 ± 0.3 hr (mean ±SEM)withmaximal firing frequencies of 4.7 ±
0.3 Hz (n = 101 neurons from 3 cultures). From the time-stamped
spikes, we found spike trains that cross-correlated either posi-
800 Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc.
tively or negatively between neurons (Figures 1B and 1C). These
cross-correlations indicated that when one neuron fired, there
was a low, but real, probability that its correlated partner
increased or decreased its discharge ratewith a short time delay.
Between-Network Validation of ConnectivityCorrelated activity can reflect functional neural connectivity
(Bialek et al., 1991; Gerstein and Perkel, 1969) but may also arise
coincidentally. To determine the likelihood of detecting spurious
versus functional connections, we developed a method (BSAC;
see Experimental Procedures) that generated an empirical distri-
bution of Z scores for false-positive connections in SCN circuits.
Iterative pair-wise analysis of spike trains from 610 neurons
Neuron
GABA Networks Destabilize the Circadian Pacemaker
recorded on 10MEAs yielded cross-correlograms of the 185,745
possible pairwise comparisons. Of these, 161,101 were impos-
sible interactions because the neurons were in physically distinct
MEAs (Figure S2). These false connections were more prevalent
than would be expected based on the standard prediction inter-
vals associated with their Z scores. This indicates that studies
that use Z scores to determine the significance of correlated
neural activity (e.g., functional neuroimaging or neural circuit
analyses) can overestimate the number of connections. By
including nodes (neurons) from independent networks (SCN cul-
tures), we were able to set an empirically derived false discovery
rate (FDR) to 0.001 (1 in every 1,000 correlations could be incor-
rect) and define functional connections as inter-neuronal firing
correlations with either jZj > 5.6 (positive cross-correlations) or
jZj > 4.68 (negative cross-correlations). Importantly, iterative
comparisons across 3–10 cultures yielded similar Z score
thresholds (p > 0.05, one-way ANOVAs for positive and negative
correlations, respectively) indicating that connection detection
was highly reproducible from culture to culture. For all SCN re-
cordings, we calculated the frequency of detecting true neuronal
interactions (hit rate) to be 96.0% ± 1.2% (mean ± SEM). Thus,
BSAC recognizes functional connections with exceptionally
high hit rates (96%) and low false-alarm rates (0.1%).
Sparse GABAergic Connectivity in the SCNWenext sought to identify the signalingmechanism(s) underlying
the identified communication between SCN neurons. Using the
significantly cross-correlated firing patterns from 330 SCN neu-
rons recorded in three cultures over 24 hr (n = 103–121 neurons/
culture), we generated spatial maps of connectivity with neurons
represented as nodes and their interactions as directed edges
(Figure 1D). We found interactions within cultures that were
inhibitory (58% ± 4%, mean ± SEM of n = 3 cultures), excitatory
(42%± 4%), or switched polarity (10%±1%) over the day (where
the proportions of the three types of interactions summed to
100% within each culture). The median lag times to maximal
cross-correlation for inhibitory and excitatory connections
were 15.4ms and 12.5ms, respectively, with 71%of correlations
peaking within 20ms and 93%within 50ms (Figure 1E). Distribu-
tions of lag times did not differ significantly between inhibitory
and excitatory connections (p = 0.24, Student’s t test). Taking
into account delays due to action potential propagation (0.1 to
6.5 ms; Figure S3) and postsynaptic response (15.4 ms as
measured by Itri et al. [2004]), these results suggest that a major-
ity of the deduced connections represent direct, fast synaptic
interactions and a minority (i.e., those with longer lag times)
may arise from polysynaptic interactions, common inputs, or
postinhibitory rebound.
Because most, if not all, SCN neurons are GABAergic and
express GABAA receptors, we tested whether the mapped
interactions depend on GABA-mediated signaling using gaba-
zine (Gbz, 100 mM) or bicuculline (200 mM). These GABAA
receptor antagonists decreased the number of significant
connections by 90% ± 2% (mean ± SEM) compared to vehicle
(p = 0.03; Figure 1F; comparing connections from 120 randomly
selected neurons in three cultures during vehicle and drug appli-
cation). This loss in functional connections was not due to a
decrease in firing since discharge rates actually increased
slightly under GABA blockade (vehicle = 4.06 ± 0.31 Hz;
mean ± SEM; GABAA-R antagonists = 5.41 ± 0.38 Hz;
p < 0.05). Those few cross-correlations that remained had low
Z-scores, suggesting that they could have been weakened by
the blockers or could depend on an alternate, weak intercellular
signal. Given the 96% hit rate of BSAC, we conclude that at least
93% of all detected interactions in these SCN cultures, both
inhibitory and excitatory, depend on GABAA receptor signaling.
Specific network topologies have been postulated to underlie
coordinated activity in a variety of biological systems including
neural networks (Grinstein and Linsker, 2005; Harris et al.,
2003; Bonifazi et al., 2009). These topologies are considered
scale-free if their degree distributions follow a power law. To
assess the architecture of the GABA-dependent network in the
SCN, we measured the number of connections from and to
each neuron (out and in node degrees, respectively). As indi-
cated by spatial network maps (e.g., Figures 1D and S2), levels
of single-cell connectivity were heterogeneous. Neurons
received connections from a median of 4.4% of the network
and sent connections to 4.5% of the network; strikingly, some
cells (15 of 330 cells) were directly connected to greater than
25% of the whole population and only 1.8% of nodes remained
unconnected. Out and in degree distributions did not differ
significantly across cultures (p > 0.05 respectively, one-way
ANOVAs) and were fit better by first-order exponential decay
functions (r2 = 0.95 and 0.87, respectively) than power functions
(out: r2 = 0.82, F(1,20) = 51.83, p < 0.0001; in: r2 = 0.81, F(1,15) =
8.13, p = 0.012, extra-sum-of-squares F test). That is, most cells
sent and received approximately 1–4 connections with exponen-
tially fewer sending or receiving progressively more connections
(Figure 1G). In these cultures (n = 3), the proportion of possible
interactions identified as functional (i.e., the network density)
ranged from 0.047 to 0.061. Together, these data suggest that
SCN neurons reliably form networks of fast neurotransmission
comprising 5% to 6% of the possible connections with patterns
that are not purely scale-free.
Because we were concerned that the density of recording
electrodes might affect the deduced topology of neural
networks, we subsampled known networks to model the effects
of undersampling and hidden nodes. We found that network
density, clustering coefficient and path length were unaffected
by including as little as 70% of the recorded neurons (Figure S4).
These results suggest that BSAC accurately revealed network
properties from recordings of 50–100 SCN neurons.
To determine if physiologically identifiable subgroups of SCN
cells were more or less connected, we linearly correlated node
degree (sending, receiving, and total interactions) with measures
of each neuron’s firing pattern at its daily peak of firing. Interest-
ingly, no metric of the interspike interval distribution (i.e., the
coefficient of variation, mode, median or mean) predicted the
degree of connectivity of single neurons. We conclude that fast
neurotransmission between SCN neurons has no apparent
preference for neurons with specific firing patterns.
VIP Signaling Does Not Alter Fast Network ArchitectureBecause VIP has been implicated in both synchronization of
circadian neurons in the SCN (Aton et al., 2005) and neural devel-
opment (Muller et al., 1995), we testedwhether VIP is required for
Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc. 801
Neuron
GABA Networks Destabilize the Circadian Pacemaker
normal GABA-dependent communication. We mapped connec-
tions within high-density, VIP null SCN cultures and found they
did not differ from wild-type cultures in network density
(0.057 ± 0.015 versus 0.045 ± 0.009, respectively; p = 0.50, n =
7 cultures per genotype), average path length (2.84 ± 0.32 nodes
versus 3.30 ± 0.23; p = 0.27), mean node degree (0.11 ± 0.03
versus 0.09 ± 0.01; p = 0.49) or mean clustering coefficient
(0.18 ± 0.03 versus 0.23 ± 0.03, respectively; p = 0.41). Together,
these data indicate that VIP signaling is not required to determine
the topology of the fast connections in the SCN. We conclude
that VIP provides a synchronizing, not a trophic, signal to coordi-
nate circadian cells within the SCN.
Connectivity Fluctuates in Circadian TimeChanges in functional connectivity over milliseconds to hours
can be critical for experience-dependent plasticity, synchroniza-
tion, or metastability in the nervous system (Harris et al., 2003).
To date, it is not known if reliable changes in functional connec-
tivity are inherent to specific synapses. To examine the dynamics
of specific connections, we monitored the strength of correlated
electrical activity from identified pairs of SCN cells over a circa-
dian cycle (Figure 2A). Iterative pair-wise analysis yielded heat
maps that convey the temporal dynamics of all connections
found within representative synchronized networks (Figure 2B).
Importantly, when both neurons fired during the day, changes
in connection strength occurred independently of circadian
changes in firing frequencies (r2 = 0.006, p = 0.27). We found
that the strength of SCN connections fluctuated (coefficient of
variation = 0.24 ± 0.01, mean ± SEM, n = 189 pairwise interac-
tions from 3 cultures), with a majority of identified connections
increasing, decreasing or oscillating in strength over the day as
opposed to varying randomly (Figure 2C).
To assess the relative stability of SCN connections, we
tracked individual, fast connections over multiple days from
three cultures. We found millisecond-level connectivity between
identified pairs of neurons that persisted over multiple days
regardless of whether their circadian patterns were in phase or
antiphase (Figures 2D and 2E, respectively). Together, these
data strongly suggest that sparse GABAergic connections can
change strength over hours, persist over multiple days within
synchronized SCN networks, and do not define a unique phase
relationship between circadian SCN neurons.
GABA Makes Daily Rhythms Less PreciseGiven that we found significant GABAA receptor-mediated inter-
actions within SCN networks that can change in strength over
time, we tested the role of these connections in modulating
circadian rhythmicity. Using a CCD camera, we monitored
single-cell rhythms in Period2::Luciferase (PER2::LUC) expres-
sion from SCN explants over 12 days with 1 to 15 min resolution
(Figure 3A). Period measurements for single cells were derived
by continuous wavelet transform analysis (CWT) and period pre-
cision was calculated based upon the variance of the continuous
period time series. Consistent with a prior report (Aton et al.,
2006), we found that GABAA receptor antagonism with 100 mM
gabazine did not alter the level (Figure S5) or average period
(23.63 ± 0.10 hr; mean ± SEM, n = 122 neurons in three SCN
explants) of cellular PER2 rhythms compared to baseline
802 Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc.
(23.42 ± 0.12 hr; p > 0.05). Importantly, GABA blockade signifi-
cantly decreased period variability of individual cells (0.70 ±
0.04 hr, mean ± SEM) compared to vehicle (0.96 ± 0.07 hr; p =
0.002; n = 3 SCN explants per treatment, 218 total cells; Fig-
ure 3B). The variability of the interpeak intervals was also
decreased during GABA blockade (0.76 ± 0.05 hr) compared
to vehicle (0.98 ± 0.08 hr, p = 0.02). Together these data show
that endogenous GABAA signaling decreases precision of circa-
dian oscillations in networked SCN neurons.
Desynchronization of Rhythmicity by GABABecause GABAA receptor signaling decreased precision of
circadian gene expression in the presence of VIP, we postulated
that GABAA receptor activation opposes the synchronizing
effects of VIP in the SCN. We monitored PER2::LUC expression
from VIP null SCN explants (Vip�/�;PER2::LUC). Consistent with
previous reports of Per1 transcription and PER2 protein in VIP-
deficient SCN (Maywood et al., 2011), circadian biolumines-
cence damped over 6 days of baseline recording (relative
amplitude error [RAE] = 0.044 ± 0.0003). SCN were then placed
in fresh media with either 100 mM gabazine or vehicle and moni-
tored for an additional 6 days (Figure 4A). Remarkably, blockade
of GABAA signaling prevented significant damping in VIP-
deficient SCN slices (for vehicle versus gabazine, RAEtreated/
RAEbaseline = 1.13 ± 0.26 versus 0.51 ± 0.10, respectively; n =
10 SCN explants per treatment; p < 0.05). CWT analysis of the
same data provided an independent quantification of the
increased amplitude of circadian rhythmicity during GABA
blockade (Figure 4B). We conclude that GABA is critical for the
loss of circadian rhythmicity in VIP-deficient SCN.
To further determine how GABAA receptor signaling impacts
circadian rhythms in single cells, we recorded bioluminescence
using a cooled-CCD camera from SCN explants. We found
that over 7 days of baseline treatment, Vip�/� cells exhibited
low amplitude PER2::LUC oscillations (RAE = 0.116 ± 0.001; Fig-
ures 4C and 4D) and progressively desynchronized (Figure 4E).
Upon medium change and addition of gabazine (100 mM), single
cell PER2 expression (Figure S6) and rhythmicity dramatically
increased (RAE = 0.004 ± 0.000, p < 0.000001 versus baseline;
n = 23 cells) and failed to damp or desynchronize for the duration
of the recording. Together these results indicate that GABAA
receptor-mediated signaling induces cycle-to-cycle jitter,
weakly opposing the stabilizing and synchronizing effects of
VIP on circadian rhythms in the SCN.
DISCUSSION
Fast Connectivity of Coupled Circadian OscillatorsIntercellular communication is necessary for proper SCN time-
keeping and regulation of daily behaviors (Yamaguchi et al.,
2003). By iteratively analyzing spike trains, we have produced
the first maps of the functional, fast connections in the SCN
network. Based on their probability to change firing within
50 ms, sensitivity to antagonists, and reproducibility across
cultures and days, we posit that at least 93% of these connec-
tions represent direct, GABAA receptor-mediated interactions
between SCN neurons. This is consistent with numerous studies
that found most, if not all, SCN neurons receive GABAergic
Figure 2. Functional Connectivity Is Dynamic over Circadian Time
(A) Circadian firing patterns with firing rate (FR) measured in Hertz (Hz) of two representative SCN neurons and the strength (S, blue circles) of their pairwise spike-
for-spike interaction every hour over 1 day. Note that the strength of this inhibitory communication increased over the day. (B) A temporal heat map demonstrates
how pairwise interactions between neurons in one SCN network changed in number and strength over 24 hr. Each row represents the strength of the connection
between a different pair of neurons every hour. Note that most interactions were inhibitory (blue) or excitatory (red) throughout the day while some switched
polarity (red and blue). (C) We categorized these representative neuron-neuron interactions as inhibitory (top) or excitatory (middle) and, using least-squares
fitting, determined whether their strength increased, decreased, oscillated, or varied randomly over the day. Pie charts (bottom) show the percentages of each
type of interaction within SCN networks (n = 189 pairwise interactions that had significant interactions for at least 4 consecutive hours from 3 cultures). Note that
the majority of inhibitory and excitatory connections changed with time of day with approximately 25% increasing in strength after the neurons started firing.
Smaller fractions of neurons either consistently decreased or increased and then decreased their interaction strength. (D and E) Connectivity persists over days
between identified neurons. Here, four representative pairs of neurons with in-phase (D) or antiphase (E) circadian firing patterns had significant communication
over multiple days. Note that an absence of strength measurements does not signify a lack of anatomic connectivity, but rather insufficient spikes to assess
activity-dependent functional connectivity.
Neuron
GABA Networks Destabilize the Circadian Pacemaker
inputs (Moore and Speh, 1993; Belenky et al., 2008). The remain-
ing 7% of connections we detected were relatively weak and
could reflect GABAergic communication incompletely blocked
by the concentrations of the antagonists used or weak signaling
via other pathways (e.g., glycine neurotransmission [Mordel
et al., 2011] or gap junctions [Long et al., 2005]). We conclude
that GABAmediates nearly all interactions capable of influencing
the millisecond firing patterns among SCN neurons.
Electron microscopic studies have found that individual SCN
neurons receive between 300–1,200 synaptic contacts (Guldner,
1976), a relatively low number compared to many brain areas.
Whole-cell recordings in acute slices report typical SCN neurons
fire during the day at 5 Hz with spontaneous IPSC frequencies at
3–12 Hz (Itri et al., 2004). Consistent with this evidence for inputs
from small numbers of neurons, we found that functional
GABAergic connectivity is sparse within SCN networks with
approximately 55% of neurons responding to inputs from 1–4
neurons. We also found that nearly every SCN neuron sends at
least one GABA-dependent output, such that most functionally
connect to approximately 5% of the network and some connect
Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc. 803
Figure 3. GABAA Receptor Signaling
Decreases Precision of Single-Cell Circa-
dian Rhythmicity
(A) Blockade of GABAA signaling with gabazine
(Gbz) reduced the variability of circadian periodi-
city in SCN cells. Traces (top) show PER2::LUC
bioluminescence from two representative cells in
SCN explants (insets, red circle shows cell loca-
tion) treated with either vehicle (left) or 100 mMGbz
(right). Heat maps (bottom) indicate the instanta-
neous amplitude of the CWT and show the stability
of the dominant period (black line) for each cell over
12 days. (B) Blockade of GABAA-dependent
signaling significantly decreased period variability
compared to vehicle (mean ± SEM; n = 120 and 98
cells, respectively; *p < 0.01).
Neuron
GABA Networks Destabilize the Circadian Pacemaker
to over 25% of the network. Taken together, these results sug-
gest GABA provides sparse, weak and fast connectivity among
SCN neurons.
Many biological networks have been hypothesized to be scale
free and follow a power-law distribution. Within the SCN, recent
computational models have predicted that a small-world scale-
free network would be the most efficient topology for circadian
synchrony (Vasalou et al., 2009; Hafner et al., 2012); however,
here we provide clear evidence that signaling through GABAA
receptors is not strictly patterned as a small-world network
and, rather than enhance synchrony, decreases the precision
and synchrony of neuronal rhythms. This is distinct from the
well-described role of GABAergic signaling in coordinating
higher-frequency (e.g., gamma [40–80 Hz]) neocortical oscilla-
tions (Ermentrout et al., 2008), but is consistent with the previous
report that GABA is not required for SCN neurons to maintain
circadian synchrony (Aton et al., 2006). In contrast to an exten-
sive literature predicting GABA-induced synchrony (Liu and
Reppert, 2000; Diekman and Forger, 2009), only a few
theoretical studies have predicted that GABA could lead to
804 Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc.
reduced precision and synchrony in neural networks (Kopell
and Ermentrout, 2004). We posit that GABAA signaling may
accelerate re-entrainment of the SCN to phase-shifted timing
cues by increasing cycle-to-cycle jitter so that the cells are
more easily phase-shifted by another signal (e.g., VIP). Our re-
sults lead us to predict that drugs that enhance GABAA signaling
(e.g., benzodiazepines) in the SCN could reduce the precision of
circadian rhythms but enhance their adjustment to new sched-
ules. Indeed, benzodiazepine administration speeds entrain-
ment of human behavioral and hormonal circadian rhythms after
simulated travel across time zones (Buxton et al., 2000).
Although it is highly unusual for GABA to be excitatory in the
adult mammalian nervous system, previous studies have pro-
vided apparently contradictory evidence that within the SCN,
GABA can be excitatory, inhibitory, or both. We found that
approximately 60% of all GABA-dependent connections were
inhibitory and 40% were excitatory throughout the circadian
day. A small fraction of these connections switched polarity for
at least one hour during the day. The differential role of inhibitory
and excitatory currents in the SCN remains unresolved.
Figure 4. Endogenous GABA Desychro-
nizes Circadian Cells in the SCN
(A) Representative traces of raw bioluminescence
from two Vip�/�;PER2::LUC SCN explants show
that gabazine (Gbz, 100 mM) rescued circadian
rhythmicity compared to vehicle.
(B) In the absence of VIP-signaling, gabazine
increased the amplitude of the dominant period
compared to baseline while vehicle treatment did
not (mean ± SEM; n = 10 explants in each condi-
tion; *p < 0.05). Total explant bioluminescence (C)
and synchrony among cells (D and E) imaged from
a Vip�/�;PER2::LUC explant (inset) in (C) quickly
damped over 6 days of baseline recording while
gabazine prevented damping of the biolumines-
cence rhythm andmaintained circadian synchrony
of the population.
Neuron
GABA Networks Destabilize the Circadian Pacemaker
Desynchronizers Opposing Synchronizers within theCircadian ClockA central focus of biological timing research has been to under-
stand the processes that promote synchronization. It is
currently appreciated that activation of Gas and Gai/o protein-
coupled receptors is vital to maintain synchrony among SCN
cells (Aton et al., 2005, 2006; Doi et al., 2011). VIP and, to a
lesser extent, vasopressin and gastrin releasing peptide have
been shown to mediate rhythm stability in and synchrony
among circadian cells (Maywood et al., 2011). In contrast, we
have found GABA to destabilize and desynchronize circadian
cells. Synchronization and desynchronization are thus both
active processes that can be differentially modulated. It is inter-
esting to speculate that developmental and seasonal changes
may alter the balance between fast neurotransmission and
slower neuropeptide signaling to adjust the timing among
SCN neurons. It is possible that GABA plays an important role
within this setting to actively drive networks of oscillators to
new phase relationships.
Additionally, recent work suggests that a hierarchy of neuro-
peptidergic signals may differentially promote or sustain rhyth-
micity and synchrony among SCN cells (Maywood et al.,
2011). In light of our results, we must place GABAA signaling
within this hierarchy and classify potential synchronizing agents
by their ability to overcome the destabilizing effects of GABA.
Our data suggest that VIP may be the only agent capable of
overcoming this destabilizing effect since it is only after VIP
signaling is eliminated that the desynchronizing effects of
GABA are unmasked. Given that VIP signaling diminishes during
aging (Cayetanot et al., 2005), increasingly unopposed
GABAergic signaling may weaken SCN neuronal synchrony
and contribute to sleep/wake cycle fragmentation in the elderly.
EXPERIMENTAL PROCEDURES
See a detailed description in the Supplemental Experimental Procedures.
Connectivity and Network Analysis
All animal procedures complied with National Institutes of Health (NIH)
guidelines and were approved by the Washington University Animal Care
and Use Committee. Spike trains from SCN neurons were recorded using
MEAs and the Z score and strength of each interaction was measured. We
graphed functional connections with GUESS software and analyzed network
architecture with NodeXL software.
BSAC (Between-Sample Analysis of Connectivity)
We developed an empirical method to discriminate correlated activity that
derived from real versus coincidental neuronal interactions. We reasoned
that correlations between spike trains recorded from neurons in physically
distinct culture dishes do not signify connectivity. Using this logic, we
iteratively cross-correlated spike trains from all neurons across 10 cultures
(samples) over 1 hr to determine the distribution of Z scores associated with
inherently false across-sample correlations. Using the full distribution of false
across-sample correlations, we determined Z scoremagnitude thresholds that
corresponded with likelihoods of discovering false-positive across-array
correlations.
Temporal Changes in Connection Strength
To determine if connection strength systematically changed with time of day,
we fit the strength versus time data with linear and cosine functions and
estimated the resultant p value using the F test.
Bioluminescence Recording and Analysis
PER2::LUC expression from SCN explants was measured using photo-
multiplier tubes or a CCD camera. Circadian period, amplitude and period
variability were assessed using FFT-NLLS andCWT. Additional statistical tests
were done with Origin 7 software. Mean periods were compared by two-tailed
t tests. Distributions of periods were compared using both Levene’s and
Brown-Forsythe’s tests for equal variance.
Drug Treatments
Bicuculline methiodide (Sigma) and SR95531 (gabazine; Tocris) were diluted
in water or DMSO. Drugs or vehicle controls were added at less than 0.5%
of total volume).
SUPPLEMENTAL INFORMATION
Supplemental Information includes seven figures and Supplemental Experi-
mental Procedures and can be found with this article online at http://dx.doi.
org/10.1016/j.neuron.2013.04.003.
ACKNOWLEDGMENTS
We thank B. Carlson, T. Holy, J. Huettner, P. Taghert, C. Zorumski, H. Herzel
and members of the Herzog lab for helpful discussions. This work was sup-
ported by NIH grants MH63104 (to E.D.H.) and F30NS070376 (to G.M.F.).
Accepted: April 2, 2013
Published: June 5, 2013
REFERENCES
Albus, H., Vansteensel, M.J., Michel, S., Block, G.D., and Meijer, J.H. (2005).
A GABAergic mechanism is necessary for coupling dissociable ventral
and dorsal regional oscillators within the circadian clock. Curr. Biol. 15,
886–893.
Aton, S.J., Colwell, C.S., Harmar, A.J., Waschek, J., and Herzog, E.D. (2005).
Vasoactive intestinal polypeptide mediates circadian rhythmicity and syn-
chrony in mammalian clock neurons. Nat. Neurosci. 8, 476–483.
Aton, S.J., Huettner, J.E., Straume, M., and Herzog, E.D. (2006). GABA and
Gi/o differentially control circadian rhythms and synchrony in clock neurons.
Proc. Natl. Acad. Sci. USA 103, 19188–19193.
Belenky, M.A., Yarom, Y., and Pickard, G.E. (2008). Heterogeneous expres-
sion of gamma-aminobutyric acid and gamma-aminobutyric acid-associated
receptors and transporters in the rat suprachiasmatic nucleus. J. Comp.
Neurol. 506, 708–732.
Bialek, W., Rieke, F., de Ruyter van Steveninck, R.R., and Warland, D. (1991).
Reading a neural code. Science 252, 1854–1857.
Bonifazi, P., Goldin, M., Picardo, M.A., Jorquera, I., Cattani, A., Bianconi, G.,
Represa, A., Ben-Ari, Y., and Cossart, R. (2009). GABAergic hub neurons
orchestrate synchrony in developing hippocampal networks. Science 326,
1419–1424.
Buxton, O.M., Copinschi, G., Van Onderbergen, A., Karrison, T.G., and Van
Cauter, E. (2000). A benzodiazepine hypnotic facilitates adaptation of
circadian rhythms and sleep-wake homeostasis to an eight hour delay shift
simulating westward jet lag. Sleep 23, 915–927.
Cayetanot, F., Bentivoglio, M., and Aujard, F. (2005). Arginine-vasopressin and
vasointestinal polypeptide rhythms in the suprachiasmatic nucleus of the
mouse lemur reveal aging-related alterations of circadian pacemaker neurons
in a non-human primate. Eur. J. Neurosci. 22, 902–910.
Choi, H.J., Lee, C.J., Schroeder, A., Kim, Y.S., Jung, S.H., Kim, J.S., Kim, Y.,
Son, E.J., Han, H.C., Hong, S.K., et al. (2008). Excitatory actions of GABA in the
suprachiasmatic nucleus. J. Neurosci. 28, 5450–5459.
De Jeu, M., and Pennartz, C. (2002). Circadian modulation of GABA function in
the rat suprachiasmatic nucleus: excitatory effects during the night phase.
J. Neurophysiol. 87, 834–844.
Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc. 805
Neuron
GABA Networks Destabilize the Circadian Pacemaker
Diekman, C.O., and Forger, D.B. (2009). Clustering predicted by an electro-
physiological model of the suprachiasmatic nucleus. J. Biol. Rhythms 24,
322–333.
Doi, M., Ishida, A., Miyake, A., Sato, M., Komatsu, R., Yamazaki, F., Kimura, I.,
Tsuchiya, S., Kori, H., Seo, K., et al. (2011). Circadian regulation of intracellular
G-protein signaling mediates intercellular synchrony and rhythmicity in the su-
prachiasmatic nucleus. Nat. Comm. 2, 327.
Ermentrout, G.B., Galan, R.F., and Urban, N.N. (2008). Reliability, synchrony
and noise. Trends Neurosci. 31, 428–434.
Gerstein, G.L., and Perkel, D.H. (1969). Simultaneously recorded trains of
action potentials: analysis and functional interpretation. Science 164,
828–830.
Grinstein, G., and Linsker, R. (2005). Synchronous neural activity in scale-free
network models versus random network models. Proc. Natl. Acad. Sci. USA
102, 9948–9953.
Guldner, F.H. (1976). Synaptology of the rat suprachiasmatic nucleus. Cell
Tissue Res. 165, 509–544.
Hafner, M., Koeppl, H., and Gonze, D. (2012). Effect of network architecture on
synchronization and entrainment properties of the circadian oscillations in the
suprachiasmatic nucleus. PLoS Comp. Biol. 8, e1002419.
Harris, K.D., Csicsvari, J., Hirase, H., Dragoi, G., and Buzsaki, G. (2003).
Organization of cell assemblies in the hippocampus. Nature 424, 552–556.
Herzog, E.D. (2007). Neurons and networks in daily rhythms. Nat. Rev.
Neurosci. 8, 790–802.
Itri, J., Michel, S., Waschek, J.A., and Colwell, C.S. (2004). Circadian rhythm in
inhibitory synaptic transmission in the mouse suprachiasmatic nucleus.
J. Neurophysiol. 92, 311–319.
Jiang, Y.J., Aerne, B.L., Smithers, L., Haddon, C., Ish-Horowicz, D., and Lewis,
J. (2000). Notch signalling and the synchronization of the somite segmentation
clock. Nature 408, 475–479.
Kopell, N., and Ermentrout, B. (2004). Chemical and electrical syn-
apses perform complementary roles in the synchronization of interneuronal
networks. Proc. Natl. Acad. Sci. USA 101, 15482–15487.
Liu, C., and Reppert, S.M. (2000). GABA synchronizes clock cells within the
suprachiasmatic circadian clock. Neuron 25, 123–128.
806 Neuron 78, 799–806, June 5, 2013 ª2013 Elsevier Inc.
Long, M.A., Jutras, M.J., Connors, B.W., and Burwell, R.D. (2005). Electrical
synapses coordinate activity in the suprachiasmatic nucleus. Nat. Neurosci.
8, 61–66.
Maywood, E.S., Chesham, J.E., O’Brien, J.A., and Hastings, M.H. (2011). A
diversity of paracrine signals sustains molecular circadian cycling in suprachi-
asmatic nucleus circuits. Proc. Natl. Acad. Sci. USA 108, 14306–14311.
Moore, R.Y., and Speh, J.C. (1993). GABA is the principal neurotransmitter of
the circadian system. Neurosci. Lett. 150, 112–116.
Mordel, J., Karnas, D., Inyushkin, A., Challet, E., Pevet, P., and Meissl, H.
(2011). Activation of glycine receptor phase-shifts the circadian rhythm in
neuronal activity in the mouse suprachiasmatic nucleus. J. Physiol. 589,
2287–2300.
Muller, J.M., Lelievre, V., Becq-Giraudon, L., andMeunier, A.C. (1995). VIP as a
cell-growth and differentiation neuromodulator role in neurodevelopment. Mol.
Neurobiol. 10, 115–134.
Strogatz, S.H. (2000). From Kuramoto to Crawford: exploring the onset of
synchronization in populations of coupled oscillators. Physica D 143, 1–20.
Vacic, V., McCarthy, S., Malhotra, D., Murray, F., Chou, H.H., Peoples, A.,
Makarov, V., Yoon, S., Bhandari, A., Corominas, R., et al. (2011).
Duplications of the neuropeptide receptor gene VIPR2 confer significant risk
for schizophrenia. Nature 471, 499–503.
Vasalou, C., Herzog, E.D., and Henson, M.A. (2009). Small-world network
models of intercellular coupling predict enhanced synchronization in the
suprachiasmatic nucleus. J. Biol. Rhythms 24, 243–254.
Wagner, S., Castel, M., Gainer, H., and Yarom, Y. (1997). GABA in themamma-
lian suprachiasmatic nucleus and its role in diurnal rhythmicity. Nature 387,
598–603.
Webb, A.B., Angelo, N., Huettner, J.E., and Herzog, E.D. (2009). Intrinsic,
nondeterministic circadian rhythm generation in identified mammalian
neurons. Proc. Natl. Acad. Sci. USA 106, 16493–16498.
Welsh, D.K., Logothetis, D.E., Meister, M., and Reppert, S.M. (1995). Individual
neurons dissociated from rat suprachiasmatic nucleus express independently
phased circadian firing rhythms. Neuron 14, 697–706.
Yamaguchi, S., Isejima, H., Matsuo, T., Okura, R., Yagita, K., Kobayashi, M.,
and Okamura, H. (2003). Synchronization of cellular clocks in the suprachias-
matic nucleus. Science 302, 1408–1412.
Neuron, Volume 78
Supplemental Information
GABA Networks Destabilize Genetic Oscillations
in the Circadian Pacemaker G. Mark Freeman Jr., Rebecca M. Krock, Sara J. Aton, Paul Thaben, and Erik D. Herzog
Supplemental Inventory
1. Supplemental Figure 1 and Legend
Demonstrates that dispersed neurons recorded on MEAs in Figure 1 maintain
similarly coordinated circadian rhythms as explant networks.
2. Supplemental Figure 2 and Legend
Demonstrates the method for empirical determination of significant, functional
neuronal connections using BSAC which are represented in Figure 1.
3. Supplemental Figure 3 and Legend
Demonstrates the conduction velocity of SCN neurons in vitro whose spike trains are
represented in Figure 1.
4. Supplemental Figure 4 and Legend
Models how properties of networks represented in Figure 1 are affected by hidden
nodes.
5. Supplemental Figure 5 and Legend
Demonstrates the effect of GABA receptor blockade on PER2::LUC expression in
SCN cells. This complements the effects of GABA receptor blockade on circadian
period as shown in Figure 3.
6. Supplemental Figure 6 and Legend
Demonstrates how GABAA receptor blockade affects peak amplitude of circadian
bioluminescence and how peak amplitude is rescued by GABA receptor blockade in
VIP-null SCN explants. This complements the effects of GABA receptor blockade on
period variability and synchrony as shown in Figures 3 and 4.
7. Supplemental Figure 7 and Legend
Illustrates how a separate analysis method validates the Z-score thresholds illustrated
in Figure 1.
8. Supplemental Experimental Procedures
9. Supplemental References
Supplemental Figure 1. Neurons in dispersed and explant networks maintain similarly
coordinated circadian rhythms. (A) Bioluminescence images from a high-density
200 µm 100 µm
dispersed SCN culture (left) and SCN explant (right). (B) Neurons within both networks
generated daily oscillations in Period2 (Per2) levels. Traces show PER2::LUC
bioluminescence from 40 randomly selected neurons in each culture. (C) The times when
SCN cells reached their peak bioluminescence (filled circles) on the fourth day of
recording were similarly distributed in these representative dispersed (left; n=174 cells)
and explant cultures (right; n = 122 cells). Arrows indicate the average phase of each
distribution. Arrow length reflects the r-value or synchronization index (SI) of each
distribution. Neurons in high-density cultures and explants are significantly synchronized
(Rayleigh test, P < 0.05, respectively). Of note, these high-density cultures had higher SI
values than what has been reported for low-density dispersed cultures in which oscillator
autonomy has been previously studied (Welsh et al., 1995; Liu et al., 2007). (D) SI values
of high-density (HD) and explant networks did not significantly differ (n = 3 cultures in
each condition, mean ± SEM; P>0.05). (E) Immunolabeling of the high-density dispersal
in S1A illustrates the network of SCN cells (blue, nuclear stained with DAPI) with
associated VIPergic somata and fibers (green).
Supplemental Figure 2. Empirical determination of significant, functional neuronal
connections using BSAC. (A) Representative examples of across-array analyses of
connectivity between 3 MEAs (left) and 7 MEAs (right). Significant correlations with
suprathreshold Z-scores (|Z| > 5.6 (positive cross-correlations) or |Z| > 4.68 (negative
cross-correlations), corresponding to a false discovery rate (FDR) of 0.001) were plotted
including within-array connections (black) and across-array connections (green). (B) The
distribution function for Z-scores of cross-correlations between SCN neurons in different
cultures determined the FDR for each Z-score. We chose a false discovery rate (FDR) of
0.001 (dashed vertical line) as our threshold to maximize the number of within-array
connections with suprathreshold Z-scores (96.0 ± 1.2%, mean ± SEM, n=10 cultures)
while minimizing the likelihood of identifying a false connection to 0.1% (1 in 1,000
connections). Importantly, at lower Z-score thresholds, the percentage of connections that
are likely true positives (hit rate, black line) decreased significantly. (C) Significant
within-array connections (black circles) have higher Z-scores and shorter latencies to the
peak of their correlation than the weaker connections with randomly distributed latencies
found among across-array connections (green circles). Z-score thresholds (dashed lines)
with a FDR=0.001 for positive (red, |Z| > 5.6) and negative (blue, |Z| > 4.68) correlations.
Sub-threshold Z-scores (grey circles) from this analysis of 10 networks (MEAs) had a
broad range of latencies. (D) We found that false connections were more frequent
(Empirical FDR) than would be predicted based on the standard prediction intervals
associated with the Z-scores (Theoretical FDR), indicating that raw Z-score statistics
(theory) can overestimate the number of connections by underestimating rates of spurious
cross-correlations. (E) Z score distributions were highly reproducible from network to
network. BSAC yielded similar Z-score thresholds for a FDR of 0.001 positive (red line)
and negative (blue) correlations regardless of whether 3 or more cultures were analyzed
(mean ± SEM).
Supplemental Figure 3. The conduction velocity of SCN neurons in vitro. (A) A
representative neuron was recorded on two electrodes 894 microns apart, i.e. somatic
signal on electrode 53 (red), and en passant recording on electrode 77 (green). Scale bar,
200 µm. (B) The extracellular potentials recorded at the two electrodes were reliably
discriminated so that (C) their spike trains were nearly identical, (D) with a 0.98 peak
probability of firing together. Based on the likelihood that these spikes arose from the
same neuron, we calculated its conduction velocity to be 0.40 m/s. These rare examples
of single neurons recorded on more than one electrode provided an estimate of the
average conduction velocity of SCN neurons (0.49 ± 0.28 m/s, mean ± SEM, n=3),
consistent with the propagation properties of unmyelinated fibers.
Supplemental Figure 4. Modeling the effects of hidden nodes on network properties.
Varying percentages of nodes from a known network (represented as 100% of the
recorded neurons) were randomly sampled 100 times, and analyzed to determine the
effects of undersampling. We found that network properties including (A) network
density, (B) average clustering coefficient and (C) average path length (mean ± SD)
changed little even when only 50% of the recorded neurons were included in the network
analysis.
0
5
10
15
20
0.00.10.20.30.40.5
012345
Net
wor
k D
ensi
ty (%
)C
lust
erin
gC
oeffi
cien
tP
ath
Leng
th (n
ode)
A
B
C
100 95 90 80 70 60 50
% of nodes sampled
0
5
10
15
20
0.00.10.20.30.40.5
012345
Net
wor
k D
ensi
ty (%
)C
lust
erin
gC
oeffi
cien
tP
ath
Leng
th (n
ode)
A
B
C
100 95 90 80 70 60 50
% of nodes sampled
Supplemental Figure 5. The effect of GABA receptor blockade on PER2::LUC
expression in SCN cells. Average peak bioluminescence over three cycles did not differ
significantly between SCN cells treated with vehicle or 100 μM Gabazine (mean ± SEM,
n = 2 cultures per treatment, n= 62 and 84 cells for vehicle and Gabazine treatment
respectively, P=0.07).
10
5
0
Bio
lum
ines
cenc
eC
ount
s/20
min
x 1
05
Vehicle Gabazine
Supplemental Figure 6. (A) GABAA receptor blockade with Gabazine (Gbz, 100 μM)
increased daily peak amplitude bioluminescence (mean ± SEM) in cells recorded from a
representative VIP-/- PER2::LUC SCN slice (same cells as in Figures 4C, 4D, 4E). (B)
Gabazine rescued the 5-day average peak amplitude of PER2 rhythms in Vip-deficient
(VIP-/-) SCN to wild-type (WT) levels (n= 23 and 344 cells, respectively; mean ± SEM).
Asterisk indicates P<0.01.
Time (d)
Bio
lum
ines
cenc
eC
ount
s/30
min
Bio
lum
ines
cenc
eC
ount
s/30
min
WT VIP-/- VIP-/-+Gbz
Bio
lum
ines
cenc
eC
ount
s/30
min
WT VIP-/- VIP-/-+Gbz
GbzA B n.s.
* *
Supplemental Figure 7. To further validate the BSAC-selected thresholds for significant
interactions between neurons (vertical dashed lines), we independently generated the
density distributions of the maximum of Z-scored variables. We randomly sampled 2% of
within-culture Z-scores, identified the maximum Z-score and repeated this 200 times to
generate the kernel density distribution of the maximum Z score (red lines) for inhibitory
(left plot) and for excitatory interactions (right plot). Comparing this distribution to the
Z-score distribution for false across-array (i.e. between SCN culture) interactions (black),
confirms that the Z-score thresholds determined by BSAC (dashed vertical gray lines)
captured greater than 99.99% of the max within-array distributions, while omitting
99.89% all Z-scores associated with false across-array interactions. These conclusions do
not change if 5% (green) or 10% (blue) of within-culture Z-scores were randomly
sampled.
0
4
8
0
4
8
1 10 100 1 10 100
Den
sity
Den
sity
Z-scoreZ-score
Inhibitory Excitatory
2%5%10%
2%5%10%
SUPPLEMENTAL EXPERIMENTAL PROCEDURES
Animals.
Homozygous Period2::Luciferase (PER2::LUC) knock-in mice (founders generously
provided by J.S. Takahashi), Vip-/- mice (founders generously provided by C.S. Colwell)
and Vip-/-;PER2::LUC mice were housed in a 12 h:12 h light:dark cycle in the Danforth
Animal Facility at Washington University. All procedures were approved by the Animal
Care and Use Committee of Washington University and followed National Institutes of
Health guidelines.
Cell Culture.
SCN were prepared as slice or high-density cultures as described previously (Webb et al.,
2009; Aton et al., 2005). For SCN explant recordings, 300 μm coronal SCN slices from
PER2::LUC mice or Vip-/-;PER2::LUC mice (age 21-28 days) were cultured at 34°C, 5%
CO2/95% air on Millicell-CM inserts (Millipore, Billerica, MA) in culture medium
(Dulbecco’s Modified Eagles Medium, DMEM, supplemented with 10% fetal bovine
serum). After four days, each explant was inverted onto a poly-D-lysine/laminin-coated
glass coverslip for 1 day prior to recording. To disperse cells, neonatal SCN (age 7-9
days) were punched from 300 μm-thick coronal brain slices and cells were dissociated
using papain. Viable cells (~10,000 cells from 10-14 SCN) were plated on either poly-D-
lysine/laminin coated multi-electrode arrays or glass coverslips and maintained in culture
medium for 3 weeks prior to recording.
Electrophysiology.
Long-term recordings of firing rate from SCN neurons were made using multi-electrode
arrays (Multichannel Systems, Reutlingen, Germany). SCN neurons were dispersed at
approximately 10,000 cells/mm2 onto sixty 30 μm-diameter electrodes (200 μm spacing)
and maintained in CO2-buffered medium for 3 weeks prior to recording. Culture
chambers fixed to arrays were covered with a fluorinated ethylenepolypropylene
membrane before transfer to the recording incubator. The recording incubator was
maintained at 36°C with 5% CO2 throughout all experiments. Extracellular voltage
signals were recorded simultaneously from 60 electrodes with 20,000 Hz sampling.
Action potentials exceeding a defined voltage threshold were digitized into 2-ms time
stamped cutouts (MC-Rack software, Multichannel Systems). Spikes from individual
neurons were discriminated offline (up to 5 neurons per electrode) based on principal
component analysis and a refractory period of greater than 2 ms (Offline Sorter, Plexon),
and firing rates were calculated over 1-min recording epochs (NeuroExplorer, Plexon).
Spike Train Analysis.
We examined the spike timing of a given target neuron during the 1 s before and 1 s after
each spike of every recorded neuron (Neuroexplorer 3.0). These cross-correlograms
included all spikes recorded over 1 h windows with 5 ms bins and were subsequently
smoothed using a 5-point boxcar filter. We confined spike train analyses to times when
both neurons fired greater than 100 spikes/hour. The Z-score of each cross-correlogram,
an indicator that a neuron increased or decreased the firing rate of a target neuron, was
measured as:
Z= | (Y-μ) / μsd |
where Y was the maximum deflection of the positive (or minimum deflection of the
negative) cross-correlation, μ was the mean of the cross-correlogram, and μsd was the
standard deviation of μ. The polarity of the cross-correlation was determined by the
polarity of the significant deflection closest to zero. The strength, S, of each
interneuronal communication was measured as:
S= | (Y- μ) / μ |
We generated graphs of functional connections between neurons with GUESS Graphical
Exploration freeware (www.graphexploration.cond.org) and analyzed network
architecture with NodeXL freeware (Microsoft, Redmond, WA).
BSAC (Between-Sample Analysis of Connectivity).
We sought an empirical method to discriminate correlated activity that derived from real
versus coincidental neuronal interactions. We reasoned that correlations between spike
trains recorded from neurons in physically distinct culture dishes do not signify
connectivity. Using this logic, we iteratively cross-correlated spike trains from all
neurons across 10 cultures (samples) over 1 h to determine the distribution of Z-scores
associated with inherently false across-sample correlations. This method has benefits
over standard techniques using scrambled or surrogate data in that no a priori
assumptions are required. Using the full distribution of false across-sample correlations,
we determined Z-score magnitude thresholds that corresponded with various likelihoods
of finding false-positive across-array correlations. Supplemental Figure 2B shows the
likelihood that a correlation of given Z-score magnitude would be found within our
across-culture dataset. For example, Z-scores above 6.0 (positive cross-correlations, red)
or 4.9 (negative cross-correlations, blue) were associated with an average of 1 connection
detected between cultures out of 2,000 comparisons, a false-discovery rate (FDR) of
0.0005. We calculated the frequency of detecting true neuronal interactions (hit rate,
black) as the difference between the number of cross-correlations with suprathreshold Z-
scores and the predicted number of false positive connections, normalized by the number
of cross-correlations with suprathreshold Z-scores. Importantly, as the Z-score thresholds
are lowered, the percentage of connections that are likely true positives (hit rate)
decreases significantly. Of note, these Z-score values did not vary significantly as a
function of the number or combination of cultures compared, indicating that these
thresholds of significance are likely valid for SCN networks generally (Supplemental
Figure 2E). We further compared the BSAC results to a published method using Granger
Causality for point processes (Kim et al., 2011) and found BSAC identified the same
connections in a set of 4 neurons, but was approximately 500 – 1000 times faster.
Additionally, comparison of the density distributions of the maximum of N Z-scored
variables to the Z-score thresholds as determined by BSAC indicate that the Z-score
thresholds as determined by BSAC capture greater than 99.99% of the max within-array
distributions, while omitting 99.89% all Z-scores associated with false across-array
interactions (Supplemental Figure 7).
Temporal change in connection strength
Traces of interaction strength were analyzed for time dependent behavior by applying
least squares fitting for a linear function
str(t) = a + b • t
and a cosine function
str(t) = m + a • cos(ω • t) + b • sin(ω • t)
For each fit, the p-value was calculated using the F-statistic (R software, www.R-
project.org). An interaction was classified as changing linearly if the p-value for the
linear fit was less than 0.05. Linear interactions were further classified as increasing or
decreasing based on the sign of the linear slope. Oscillating interactions were not
significantly fit by a linear function, and had a p-value for the cosine fit less than 0.05.
Random interactions include all connections not grouped into the above categories.
Bioluminescence recording and analysis.
SCN explants were placed in sealed 35-mm Petri dishes (BD Biosciences, San Jose, CA)
with pre-warmed, recording medium (DMEM supplemented with 10 mM HEPES and
100 μM beetle luciferin (Promega, Madison, WI). Bioluminescence was recorded at 36oC
in air using a photomultiplier tube (PMT) (HC135-11 MOD, Hamamatsu Corp.,
Shizuoka, Japan) as described (Aton et al., 2006). Bioluminescence counts were
integrated and stored at 1-min intervals for up to 12 days. For single-cell recordings, SCN
explants or high-density dispersals were incubated in recording medium and imaged
using a Stanford XR/Mega-10Z CCD (Stanford Photonics) or Versarray 1024 cooled-
CCD camera (Princeton Instruments) at 36oC in air. Photon counts were integrated every
one to fifteen minutes, respectively, processed using 2 x 2 spatial binning and quantified
using Image J software.
We assessed circadian period and amplitude of gene expression from SCN
explants and single-cells using FFT-NLLS (Plautz et al., 1997) and wavelet analysis
(Meeker et al., 2011) from at least 4 days of bioluminescence. We quantified period
variability by the relative amplitude error (RAE) of the FFT-NLLS, the standard
deviation of the continuous period measurement of the continuous wavelet analysis and
by peak-to-peak timing. Additional statistical tests were done with Origin 7 software
(OriginLab, Northampton, MA) and Prism 4 software (Graphpad , La Jolla, CA). Mean
periods were compared by two-tailed t-tests. Distributions of periods were compared
using both Levene's and Brown-Forsythe's tests for equal variance.
Immunostaining.
SCN dispersals on glass coverslips were treated with colchicine (Sigma; 50 g/ml)
overnight before staining. Cells were fixed using 4% paraformaldehyde, 4% sucrose and
immunolabeled for neuropeptides using standard methods (Webb et al., 2009; Aton et al.,
2005). Cultures were labeled with rabbit anti-VIP (Immunostar; 1:2,000) followed by
donkey anti–rabbit IgG conjugated to Cy2 (Jackson ImmunoResearch; 1:50). Nuclei were
visualized using DAPI staining.
Drug Treatments.
Bicuculline methiodide (Sigma, Saint Louis, MO) and SR95531 (gabazine; Tocris,
Ellisville, MO) were diluted in deionized water or DMSO. Drugs or vehicle controls were
added to the 1 ml of recording medium (<0.5% of total volume).
SUPPLEMENTAL REFERENCES
Kim, S., Putrino, D., Ghosh, S. and Brown, E.N. (2011). A Granger causality measure for
point process models of ensemble neural spiking activity. PLoS Comp. Biol. 7,
e1001110.
Meeker K., Harang R., Webb A.B., Welsh D.K., Doyle F.J. 3rd, Bonnet G., Herzog E.D.,
and Petzold L.R. (2011). Wavelet measurement suggests cause of period instability in
mammalian circadian neurons. J. Biol. Rhythms 26, 353-62.
Plautz J.D., Straume M., Stanewsky R., Jamison C.F., Brandes C., Dowse H.B., Hall J.C.,
and Kay S.A. (1997). Quantitative analysis of Drosophila period gene transcription in
living animals. J Biol Rhythms 12, 204-217.
top related