Article

A Central Amygdala-Substantia Innominata Neural

Circuitry Encodes Aversive Reinforcement Signals
Graphical Abstract Authors
Yuting Cui, Guanghui Lv, Sen Jin, ...,
Fuqiang Xu, Tonghui Xu, Haohong Li

Correspondence
xutonghui@hust.edu.cn (T.X.),
hxli@hust.edu.cn (H.L.)

In Brief
Cui et al. show that central amygdala
PKC-d+ neurons can modulate negative
reinforcement learning by transmitting
aversive signals to the substantia
innominata.

Highlights
d CeL PKC-d+ neurons encode aversive signals in negative
reinforcement learning

d The CeLPKC-d+-SI neural circuit modulates negative

reinforcement learning

d SI glutamatergic neurons are responsible for receiving

aversive signals from the CeL

Cui et al., 2017, Cell Reports 21, 1770–1782

November 14, 2017 ª 2017 The Author(s).
https://doi.org/10.1016/j.celrep.2017.10.062
 Cell Reports

Article

A Central Amygdala-Substantia Innominata Neural

Circuitry Encodes Aversive Reinforcement Signals
Yuting Cui,1,2 Guanghui Lv,1,2 Sen Jin,3 Jie Peng,1,2 Jing Yuan,1,2 Xiaobin He,3 Hui Gong,1,2 Fuqiang Xu,1,2,3
Tonghui Xu,1,2,* and Haohong Li1,2,4,*
1Britton Chance Center for Biomedical Photonics, Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and
Technology, Wuhan, Hubei 430074, China
2MoE Key Laboratory for Biomedical Photonics, Collaborative Innovation Center for Biomedical Engineering, School of Engineering Sciences,

Huazhong University of Science and Technology, Wuhan, Hubei 430074, China

3Center for Brain Science, Key Laboratory of Magnetic Resonance in Biological Systems and State Key Laboratory of Magnetic Resonance

and Atomic and Molecular Physics, Wuhan Institute of Physics and Mathematics, CAS Center for Excellence in Brain Science and Intelligence
Technology, Chinese Academy of Sciences Wuhan, China
4Lead Contact

*Correspondence: xutonghui@hust.edu.cn (T.X.), hxli@hust.edu.cn (H.L.)

https://doi.org/10.1016/j.celrep.2017.10.062

SUMMARY aversive reinforcers remains a major endeavor of behavioral

Aversive stimuli can impact motivation and support
associative learning as reinforcers. However, the
neural circuitry underlying the processing of aversive
reinforcers has not been elucidated. Here, we report
that a subpopulation of central amygdala (CeA)
GABAergic neurons expressing protein kinase
C-delta (PKC-d+) displays robust responses to aver-
sive stimuli during negative reinforcement learning.
Importantly, projections from PKC-d+ neurons of
the CeA to the substantia innominata (SI) could
bi-directionally modulate negative reinforcement
learning. Moreover, consistent with the idea that
SI-projecting PKC-d+ neurons of the CeA encode
aversive information, optogenetic activation of this
pathway produces conditioned place aversion, a
behavior prevented by simultaneous ablating of SI
glutamatergic neurons. Taken together, our data
define a cell-type-specific neural circuitry modu-
lating associative learning by encoding aversive
reinforcement signals.
INTRODUCTION
Associative learning is typified by associating a conditioned
stimulus (CS) with a reinforcer (unconditioned stimulus [US]),
such as appetitive or aversive reinforced stimuli, to guide
learning (Fanselow and Poulos, 2005; LeDoux, 2000). In complex
environments, associating danger with conditioned or predictive
danger cues is important in animals for the rapid avoidance of
threats. Negatively reinforced associative learning, which is a
form of operant conditioning that allows organisms to orient
toward specific goals in the environment and flexibly control
actions, has been documented behaviorally (Avarguès-Weber
et al., 2010; Cybulska-Klosowicz et al., 2009; O’Doherty et al.,
2004; Resnik et al., 2012). However, determining the precise
neuronal circuits and mechanisms underlying the processing of
neuroscience. An impressive body of literature suggests that
the amygdala complex is critical for the integration and control
of classical fear conditioning in which animals learn to associate
aversive events with stimulus cues (Ehrlich et al., 2009; Hauben-
sak et al., 2010; Herry et al., 2008). Thus, one candidate structure
for negatively reinforced associative learning is the central amyg-
dala (CeA), which contains both medial (CeM) and lateral (CeL)
divisions and serves as a major output of the amygdala complex
(Davis et al., 2003; Ehrlich et al., 2009; Paré et al., 2004). There
are two major subpopulations of GABAergic neurons found in
the CeL, one expressing the neuropeptide somatostatin (SOM)
and the other expressing protein kinase C-delta (PKC-d) (Hau-
bensak et al., 2010; Li et al., 2013). These two subpopulations
have distinct functions in associative fear learning (Ciocchi
et al., 2010; Haubensak et al., 2010; Li et al., 2013). SOM neurons
represent CeLon neurons (excited in response to CS) and PKC-d
neurons represent CeLoff neurons (inhibited in response to CS)
during fear expression (Ciocchi et al., 2010; Fadok et al., 2017;
Haubensak et al., 2010; Li et al., 2013; Yu et al., 2016). Previous
research has revealed that the firing rates of CeA neurons
increase selectively at the time of reward omission in well-trained
rats during a forced-choice task (Calu et al., 2010). Amygdala
responses could be evoked by aversive unconditioned foot-
shock stimuli (McHugh et al., 2014). In addition, the parabrachial
nucleus, as an upstream brain region, selectively transmits the
US aversive signal to the CeL during aversive learning (Han
et al., 2015). However, it remains to be determined which cell
types in the CeL encode aversive signals.
Previous research has shown that CeA neurons project prom-
inently to and form symmetric synapses mainly in the subnuclei
ventrolateral substantia innominata (SI) of the basal forebrain
(BF) (Jolkkonen et al., 2002). BF cholinergic neurons show short
latency activation after the delivery of punishment in an auditory
detection task and convey precisely time-locked signals to large
areas of the brain (Hangya et al., 2015). Two other populations in
the BF, GABAergic and glutamatergic neurons, have been impli-
cated in encoding motivational salience in go/no-go associative
learning (Avila and Lin, 2014; Hangya et al., 2015; Lin and Nico-
lelis, 2008). Interestingly, in a similar behavioral task, in vivo

1770 Cell Reports 21, 1770–1782, November 14, 2017 ª 2017 The Author(s).
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
 (legend on next page)

Cell Reports 21, 1770–1782, November 14, 2017 1771

microendoscopic calcium imaging showed that these three cell
types could be activated by overt punishment stimuli (Harrison
et al., 2016). These results suggest that neuronal activity in the
BF is highly heterogeneous and that the definition of cell types
in the BF are relatively imprecise. A way to solve this imprecision
is to account for the input and output of BF neurons with specific
function. It has been proposed that the CeA, via its influence in
the BF, might function to enhance the gain in appetitive or aver-
sive stimuli involved in associative learning (Gozzi et al., 2010;
Pape, 2010; Peck and Salzman, 2014). However, the BF-projec-
ting specific cell types in the amygdala and the precise neural
circuit involved in associative learning are unknown.
Here, we investigated the role of CeLPKC-d+ and CeLSOM+ neu-
rons in a go/no-go associative learning task. Fiber photometry re-
vealed that the CeLPKC-d+ neurons, but not the CeLSOM+ neurons,
represented an aversive signal during negative reinforcement
learning of the go/no-go task. Circuitry mapping and functional
manipulation suggested that the SI-projecting CeLPKC-d+ neurons
could bi-directionally modulate negative reinforcement learning.
Further analysis found that these neurons recruited SI glutama-
tergic neurons (Vglut2) to drive conditioned place aversion
(CPA). Together, our data uncovered a previously unknown func-
tion of the CeLPKC-d+-SIVgluT2+ neural circuitry to modulate nega-
tive reinforcement learning by encoding aversive signals.
RESULTS
Role of CeLPKC-d+ Neurons in the Negative
Reinforcement Learning
As reported in previous work, PKC-d and SOM account for
90% of all CeL neurons and are largely non-overlapping (Hau-
bensak et al., 2010; Li et al., 2013). To identify the neuronal activ-
ity patterns of these two subpopulations in associative learning,
we trained head-fixed mice to perform a go/no-go task, which is
a well-established behavioral paradigm used to study associa-
tive learning by associating CS with reinforcer reward or punish-
ment (US), respectively (Figures 1A and 1B). The go cue requires
mice to lick within a response time window to acquire a reward
(appetitive learning), whereas the no-go cue requires mice to
withhold their responses to avoid punishment (negative rein-
forcement learning) (Dolzani et al., 2013; Gomez et al., 2007).
During training, the performance (Figure 1C) and correct rejec-
tion rate (Figures S1A and S1B) gradually increased within the
initial 5–6 sessions, after which performance reached a stable
level, but there was a ceiling effect for the hit rate (Figures S1A
and S1B) across learning phases. The well-trained mice could
discriminate between go and no-go cues (Figure S1C).
We used fiber photometry, which is a method for collecting
population intracellular [Ca2+] fluorescent signals from a geneti-
cally encoded Ca2+ indicator through a single chronic fiber optic
implant (Akerboom et al., 2012; Chen et al., 2013), to monitor
CeLPKC-d+ and CeLSOM+ neuronal activities during a go/no-go
task. A Cre-dependent adeno-associated virus (AAV) encoding
Ca2+ indicator GCaMP5g (AAV2-EF1a-DIO-GCaMP5g) or
AAV2-EF1a-Flex-GFP (control) was injected into the CeL of
transgenic PKC-d-Cre mice, and an optical fiber was then
implanted above the CeL for simultaneous delivery of a
473-nm excitation laser and collection of GCaMP5g fluores-
cence emissions (Figure 1D). GCaMP5g expression was largely
limited to the CeL, with minor expression in the CeM (Figure 1D).
Immunofluorescence staining confirmed the high specificity of
GCaMP5g expression in the CeLPKC-d+ neurons (mean ± SEM,
88.8% ± 2.1%; n = 3 mice; Figure 1E). In early associative
learning, the CeLPKC-d+ neurons showed time-locked increases
during the air puff stimulus (average peak DF/F ± SEM,
11.67% ± 2.14% ) in negative reinforcement learning and peak
DF/F without statistical significance was observed on reward
port entries (water) in appetitive learning (Figures 1F and 1G).
Similarly, during late learning (Figure 1H) and expression of
learning (Figure 1I), the punishment event was still reliably asso-
ciated with a significant increase in GCaMP5g signals in the
CeLPKC-d+ neurons (late phase: average peak DF/F ± SEM,
9.06% ± 1.84%; expression phase: average peak DF/F ± SEM,
7.98% ± 1.14%). We observed slight, non-significant suppres-
sion in the activity of CeLPKC-d+ neurons in response to water
(Figures 1H and 1I). US-evoked responses were not observed
in the PKC-d-GFP control mice during the training phase (Fig-
ure 1J). We next tracked the activity of CeLSOM+ neurons in
response to punishment or reward in the go/no-go task
(Figures S2A, S2D, and S2E). AAV2-EF1a-DIO-GCaMP5g was
injected into the CeL of Som-IRES-Cre mice, with an optic fiber
implanted above the CeL (Figures S2B and S2C). We found that

Figure 1. Aversive Stimulus Increases Ca2+ Signals of CeLPKC-d+ Neurons in Negative Reinforcement Learning
(A) Behavioral setup of the go/no-go task and definition of four behavioral outcomes.
(B) Paradigm of the go/no-go task. A tone (2 clicks/s) was used as the go cue. A light or tone (50 clicks/s) was used as the no-go cue to counterbalance the
modality-specific activity of CeLPKC-d+ neurons.
(C) Performances in the go/no-go task in 7 PKC-d-Cre mice (4 PKC-d-CreGCaMP5g mice and 3 PKC-d-CreGFP mice; light was used as the no-go cue) and 3 PKC-d-
CreGCaMP5g mice (tone was used as the no-go cue). Data are presented as means ± SEM.
(D) Schematic of GCaMP5g viral targeting to CeLPKC-d+ neurons (left). Confocal image of GCaMP5g expression in the CeL and location of optical fiber (right).
Scale bar, 100 mm.
(E) Top: representative images of GCaMP5g (green) and cell-type-specific markers (red: PKC-d+ neurons) expressed in the CeL. Scale bars, 100 mm. Bottom:
enlargements of boxed areas. Scale bar, 10 mm.
(F) Heatmap of Ca2+ signals from mouse aligned to the initiation of water and air puff during the early phase of learning. Color scales on the right indicate DF/F.
(G I) Mean Ca2+ transients from seven PKC-d-CreGCaMP5g mice aligned to events starting with water and air puff during early learning (G), late learning (H), and
expression of learning (I) (early phase of learning, 1–3 sessions; late phase of learning, 6 8 sessions; expression of learning, 9 11 sessions). **p = 0.002 (G);
**p = 0.002 (H); **p = 0.003 (I).
(J) Mean Ca2+ transients from three PKC-d-CreGFP mice aligned to events starting with air puff and water across whole phases.
A Mann-Whitney U test was used to identify differences between baseline (average of 0.5 s before cues) and mean peak values (average duration of water or air
puff delivered). Thick lines indicate mean, and shaded areas indicate SEM.
See also Figure S1.

1772 Cell Reports 21, 1770–1782, November 14, 2017

Figure 2. Inhibition of CeLPKC-d+ Neurons Impairs Negative Reinforcement Learning
(A) Bilateral stereotaxic delivery of virus and implantation of optical fibers into the CeL of PKC-d-Cre mice.
(B) Confocal image of a coronal brain section taken from a PKC-d-Cre mouse bilaterally injected with AAV-EF1a-Flex-ArchT-GFP in the CeL. Scale bar, 500 mm.
(C) Experimental timeline.
(D) Paradigm of go/no-go task.
(E–H) Compared with the control group (n = 8), ArchT-expressing mice (n = 8) showed significantly lower performance (E), discriminability (F), and correct-
rejection rates (H), but not hit rates (G). Two-way analysis of variance with mixed design; F(1, 14) = 12.48, **p = 0.003 (E); F(1, 14) = 6.41, *p = 0.024 (F); F(1, 14) =
9.28, **p = 0.009, n.s., not significant (H). Data represent means ± SEM.
(I) Distribution of the response delay in go trials.
(J) Median response delay for each group.
(K and L) Performance (K) and correct-rejection (L) rate of mice in the last session of learning without laser (training) and with laser (expression). n.s., not significant
(Mann-Whitney U test). Boxplots show the mean (solid square), median, quartiles (boxes), and range (whiskers).

no matter in the early learning, the late learning, or learning
expression, there were no significant changes in GCaMP5g fluo-
rescence in SOM+ neurons following punishment, reward port
entries (Figures S2F–S2I). In addition, across the whole training
phases, SOM+ neurons had no response to an aversive condi-
tioned stimulus (Figures S2J). The above data illuminate the
dynamics of CeLPKC-d+ and CeLSOM+ neurons during associative
learning, demonstrating that CeLPKC-d+ neurons, but not SOM+
neurons, encode aversive signals during negative reinforcement
learning.
To examine whether PKC-d+ neurons in the CeL are involved in
negatively reinforced associative learning, we bilaterally injected
AAV encoding Cre-inducible archaerhodopsin (ArchT), a light-
sensitive inhibitory proton pump (AAV9-EF1a-Flex-ArchT-GFP),
or control virus (AAV2-EF1a-DIO-EYFP) into the CeL of PKC-d-
Cre mice and implanted optic fibers bilaterally into the CeL (Fig-
ures 2A and 2B). A laser was delivered to activate ArchT in vivo,
thereby inhibiting CeLPKC-d+ neurons in a cell-type-specific
manner. Mice were trained to perform the go/no-go behavioral
task (Figures 2C and 2D). We found that the ArchT mice with
photoinhibition of PKC-d+ neurons exhibited significantly lower
performance across the entire course of training (Figure 2E).
Detailed analysis revealed that PKC-d+ArchT mice showed lower
discriminability (Figure 2F) (used to evaluate an animal’s ability
to distinguish between two stimuli) and a poorer correct-rejection
rate (Figure 2H) in the no-go trials. The hit rates in the go trials

Cell Reports 21, 1770–1782, November 14, 2017 1773

were similar between the two groups (Figure 2G). In addition,
photoinhibition of PKC-d+ neurons did not influence the speed
of decision-making in learning, as assessed by reaction delay
(Figures 2I and 2J). Furthermore, laser illumination did not impair
the behavioral performance or correct rejection rate during the
expression of associative learning (Figures 2K and 2L), suggest-
ing that CeL PKC-d+ neurons could modulate the acquisition, but
not the expression, of negative reinforcement learning.
Circuitry Mechanism of CeLPKC-d+ Neurons Encoding
Aversive Signals
To elucidate the circuitry mechanism through which PKC-d+
neurons convey aversive signals, we mapped the whole-brain
targets of CeLPKC-d+ neurons in combination with fluorescence
micro-optical sectioning tomography (fMOST) (Li et al., 2010;
Xu et al., 2013). AAV2-EF1a-DIO-EYFP was injected into the
CeL of PKC-d-Cre mice. 21 days later, the brain was embedded,
continuously sliced, and imaged by the fMOST system. Massive
axonal fibers were observed in the ventral bed nucleus of the
stria terminalis (BNSTv), BF, CeM, retrorubral field (RRF), and
lateral parabrachial nucleus (LPB) (Figure S3). Considering the
functions of the BF underlying associative learning (Avila and
Lin, 2014; Hangya et al., 2015; Harrison et al., 2016; Lau and
Salzman, 2008), we next focused on the study of the CeL-BF cir-
cuit. Confocal imaging revealed that prominent terminals of the
CeLPKC-d+ neurons (minor terminals of the CeMPKC-d+ neurons)
projected to the BF, particularly in the SI (Figures 3A and 3B).
To further determine whether the CeL neurons formed monosyn-
aptic connections with neurons in the SI, we performed Cre-
dependent rabies-virus-based transsynaptic retrograde tracing
(Miyamichi et al., 2011; Wall et al., 2010; Wickersham et al.,
2007). It is known that cholinergic (ChAT+) neurons comprise
5% of total neurons in the BF, whereas GABAergic (GABA+)
neurons account for 35%, and others, including glutamatergic
neurons, account for 60% (Henny and Jones, 2008; Mayo
et al., 1984). We co-injected helper viruses encoding Cre-depen-
dent avian sarcoma leucosis virus glycoprotein EnvA receptor
(TVA) (AAV5-CAG-FLEX TVA66T-mCherry) and rabies virus
envelope glycoprotein (RG) (AAV5-CAG-FLEX-RG) into the SI
of VgluT2-Cre, GAD2-Cre, or ChAT-Cre mice. 3 weeks later,
the EnvA-coated rabies-YFP virus (DRG-YFP) was then injected
into the same place (Figure 3C). Starter cells expressing both
mCherry and YFP were observed in the SI (Figure 3D). Retro-
gradely labeled (YFP+) cells were observed in the CeL (Fig-
ure 3D), which confirmed that CeL neurons could form monosyn-
aptic connections with the three cell types in the SI. The CeL-SI
connection was further characterized by immunostaining of the
PKC-d protein in the CeL, with retrograde tracer cholera-toxin
B (CTB) injected into the SI (Figure 3E). Results showed that
84.55% ± 2.35% of CTB-labeled neurons in the CeL overlapped
with PKC-d+ neurons (30.28% ± 5.9% of PKC-d+ neurons in the
CeL projected to the SI; Figure 3G). Furthermore, we tested
whether neurons expressing SOM projected to the SI by inject-
ing CTB-647 into the SI of Som-IRES-Cre; Ai3 mice and found
that a few SOM+ neurons in the CeL projected to the SI
(5.28% ± 0.56% of CTB-labeled neurons expressing SOM+ neu-
rons; Figures 3F and 3G). These results demonstrate the contri-
bution of CeLPKC-d+ neurons to the SI-projecting cell population.
1774 Cell Reports 21, 1770–1782, November 14, 2017
Considering CeLPKC-d+ neurons displayed robust responses to
aversive stimulus (Figure 1), we next sought to examine whether
the activation of the CeLPKC-d+-SI circuit was sufficient to facili-
tate negative associative learning as an aversive reinforcer.
Accordingly, AAV encoding Cre-inducible channelrhodopsin-2
(ChR2), direct light-activated cation-selective ion channel
(AAV5-EF1a-DIO-hChR2-EYFP), or control virus (AAV2-EF1a-
DIO-EYFP) was injected into the CeL of PKC-d-Cre mice, with
an optical fiber implanted above the SI (Figures 4A and 4B).
Mice were then trained to perform the go/no-go task (Figure 4C).
In the task, mice were divided into three groups that received two
different aversive reinforced stimuli or no stimulus (Figure 4C).
Optogenetic activation of CeLPKC-d+-SI ChR2-expressing termi-
nals accelerated the learning rate in the group in which the air
puff was replaced with a 473-nm laser (2 s, 30 Hz, 10 ms)
compared with the other two groups (Figure 4D). The effects
on behavior were caused mainly by an increase in the correct
rejection rate of negative reinforcement learning (Figure 4G),
without significant changes in discriminability and the hit rate
of appetitive learning (Figures 4E and 4F). The ChR2-expressing
mice also performed better than the control mice trained with an
air puff as the aversive reinforced stimulus, which was related to
the higher correct rejection rate in the initial two sessions (Fig-
ure 4G). These results indicate that the effects caused by activa-
tion of the CeLPKC-d+-SI circuit were similar to the behavioral
effects induced by air puff as an aversive reinforced stimulus,
which coincided with the activity pattern of CeLPKC-d+ neurons
(Figure 1). The above data raised the question of whether
silencing the circuit would impair negative reinforcement
learning. We inhibited the activity of the CeLPKC-d+-SI circuit by
injecting AAV9-EF1a-Flex-ArchT-GFP or control virus (AAV2-
EF1a-DIO-EYFP) into the CeL of PKC-d-Cre mice, with an optical
fiber implanted above the SI (Figures 4H and 4I). Mice were
trained to perform the go/no-go task in accordance with previ-
ous experimental procedures (Figures 2C and 2D). Indeed,
specific inhibition of the CeLPKC-d+-SI projections impaired
acquisition of negative reinforcement learning similar to the
effects of inhibition of PKC-d+ neurons, as indicated by the lower
discriminability and lower correct rejection rate in the no-go tri-
als, but maintained the hit rate in the go trials (Figures 4J–4M).
The reaction delay was not altered (Figures 4N and 4O). Further-
more, inhibiting the CeLPKC-d+-SI projections did not interfere
with the expression of negative reinforcement learning (Figures
4P and 4Q). To further inhibit activity of the CeLPKC-d+-SI circuit
in behavioral outcomes, we applied continuous laser illumination
(532 nm) during the behavioral outcomes (Figure S4A). ArchT-
expressing mice indeed showed poorer performance and
reduced discriminability and correct-rejection rates, but not hit
rates, compared with mice in the control group (Figures
S4B–S4E), which were similar to the effects of inhibiting
CeLPKC-d+-SI circuit across the entire period from the start of
the trials to the end of the response window in all trials (Figure 4).
Together, these results demonstrate that the CeLPKC-d+-SI circuit
could bi-directionally modulate negative reinforcement learning
in the go/no-go task.
As the CeLPKC-d+-SI neural circuitry transmits an aversive
signal, we sought to test whether activation of the
circuitry could drive aversive-related behavior. We injected
Figure 3. CeLPKC-d+ Neurons Project to the SI
(A) Virus expression in CeL of PKC-d-Cre mice. Scale bars represent 500 mm (left) and 100 mm (right). Approximate distance from the bregma is shown at the
bottom.
(B) Representative histological images of the SI innervated by axonal terminals from CeLPKC-d+ neurons. Scale bars represent 500 mm (left) and 200 mm (right).
Similar results were confirmed in two other mice. Approximate distance from the bregma is shown at the bottom. CPu, caudate putamen (striatum); LGP, lateral
globus pallidus; LH, lateral hypothalamic area.
(C) Virus injection procedures for rabies retrograde tracing. Helper viruses: AAV5-CAG-FLEX TVA66T-mCherry, AAV5-CAG-FLEX-RG (1:1). Rabies virus: EnvA-
RV-YFP.
(D) Typical monosynaptic rabies tracing coronal sections of the anatomical localization of the BF and CeL. Scale bars represent 200 mm (BF), 20 mm (starter
neuron), and 50 mm (CeL). The results were repeated in at least three mice per group.
(E) Top left: coronal brain section showing the SI injected with CTB-488 (green), scale bar, 500 mm. Top right: projecting neurons labeled with CTB (green). PKC-d+
neurons stained with Alexa Fluor 594 (red). Scale bar, 50 mm. Bottom: higher-magnification of images of the noted region in the CeL. Scale bar, 10 mm.
(F) Confocal images of a coronal brain section from a Som-IRES-Cre; Ai3 mouse with CTB-647 injected into the SI. Scale bars represent 100 mm (top) and 10 mm
(bottom, higher-magnification images).
(G) Statistical graphs showing co-labeling of CTB-labeled neurons and PKC-d+ or SOM+ neurons in the CeL. n = 3 mice/group. Boxplots show mean (solid
square), median, quartiles (boxes), and range (whiskers).
See also Figure S3.

AAV5-EF1a-DIO-hChR2-EYFP or control virus (AAV2-EF1a-
DIO-EYFP) into the CeL of PKC-d-Cre mice and tested the con-
sequences of optogenetically activating SI fibers from CeLPKC-d+
neurons in a conditioned place preference/avoidance (CPP/A)
assay. Using a revised 3-day protocol, ChR2- and EYFP-ex-
pressing (control) mice were tested in a two-chamber apparatus
(Figure 5A). On the conditioning day, laser stimulation (30 Hz,
10 ms) was randomly paired to one of the two chambers and
delivered whenever the mice entered the paired chamber.
Mice were tested twice more on day 3 (Figure 5A). On post-
test 1 without the laser, the ChR2CeL-SI terminals photostimu-
lated on day 2 mice displayed obvious CPA, whereas control

Cell Reports 21, 1770–1782, November 14, 2017 1775

Figure 4. CeLPKC-d+-SI Projections Modulate Negative Reinforcement Learning
(A) Schematic of the experimental approach.
(B) Top: confocal image of a coronal brain section taken from a PKC-d-Cre mouse with AAV5-EF1a-DIO-hChR2-EYFP injected into the CeL. Scale bar, 200 mm.
Bottom: ChR2-expresssion terminals were observed in the SI. Scale bar, 200 mm.
(C) Experimental design of the go/no-go task.
(D–G) Performance (D), discriminability (E), hit rate (F), and correct-rejection rate (G) of the three groups. Black line represents comparison of the ChR2-expressing
group with laser as punishment (n = 6) and the control group with laser as punishment (n = 7); green line represents comparison of the ChR2-expressing group with
laser as punishment and the control group with air puff as punishment (n = 6). n.s., not significant (two-way ANOVA with mixed design). (D) Black line:
(legend continued on next page)

1776 Cell Reports 21, 1770–1782, November 14, 2017

 Figure 5. Activation of CeLPKC-d+-SI Projec-
tions Drives CPA
(A) Procedure of CPP/A performance.
(B) Traces on day 1 (pre-test) and day 3 (post-test 1
and post-test 2).
(C) Post-test 1/pre-test and post-test 2/pre-test
ratios of time spent in conditioned chamber.
CeLPKC-d+-SI-GFP (black), n = 7; CeLPKC-d+-SI-
ChR2 (blue), n = 7. Post-test 1, *p = 0.035; post-
test 2, **p = 0.002 (Mann-Whitney U test).
(D) Time spent in conditioned chamber on day 3
and day 1. Post-test 1: *p = 0.018; post-test 2: **p =
0.002 (Mann-Whitney U test). Boxplots show mean
(solid square), median, quartiles (boxes), and range
(whiskers) .

mice spent a similar amount of time in the two chambers (Figures
5B–5D). Moreover, on post-test 2, real-time stimulation of
CeLPKC-d+ neuronal terminals whenever mice entered the condi-
tioned chamber on day 2 caused a further increase in CPA (Fig-
ures 5B–5D). These data show that activation of the CeLPKC-d+-SI
circuit drives CPA by transmitting an aversive signal.
SI VgluT2+ Neurons Are Responsible for Receiving an
Aversive Signal from CeLPKC-d+ Neurons
Although enhanced activity of the CeLPKC-d+-SI circuit drove
CPA, the subpopulation of SI neurons receiving aversive signals
from the CeLPKC-d+ neurons was unknown. To address this, we
first recorded activity of the SI neurons when CeLPKC-d+-SI pro-
jection terminals were activated (Figure S5A). A total of 53 single
units were recorded using a moveable, 32-channel optetrode
microwire bundle in three mice in which electrode placements
were confirmed in the SI (Figure S5B). We observed that the firing
rates of 13.21% neurons (7/53 neurons; Figures S5C and S5D)
were increased when the activity of the CeLPKC-d+-SI projection
terminals was enhanced. The excited neurons exhibited a long
latency time following laser stimulation, indicating that informa-
tion was transmitted across SI multi-synaptic pathways (Ciocchi
et al., 2015). However, excited neural types in the SI were
unknown. Neuronal heterogeneity in the BF is an important chal-
lenge in the study of such neurons (Harrison et al., 2016). To visu-
alize the three spatially intermingled major cell types in the SI, we
injected AAV9-CAG-ChR2-GFP into the CeL of ChAT-YFP mice
or VgluT2-Cre, GAD2-Cre, and ChAT-Cre mice in which
VgluT2+, GABA+, and ChAT+ neurons were respectively labeled
by injecting AAV2-EF1a-Flex-GFP into the SI (Figures S6A and
S6B). Compared with the control condition (no laser), photoacti-
vation of the CeLPKC-d+ neuronal fibers induced c-Fos expres-
sion, a surrogate marker for neuronal activation, in SI VgluT2+
(Figures S6C and S6D) and ChAT+ neurons (Figures S6G and
S6H), but not in GABA+ neurons (Figures S6E and S6F). These
results indicate that VgluT2+ and ChAT+ neurons are more likely
to receive an aversive signal, although the possible involvement
of GABA+ neurons and c-Fos expressed in a cell-type-specific
way could not be ruled out. In order to validate it from the
behavior, we optogenetically activated the three types of neu-
rons in a real-time place preference/aversion task (RTPP/A) (Fig-
ure 6A). Only activation of the VgluT2+ neurons induced CPA
(Figures 6B–6D). Activation of the GABA+ neurons induced a
preference in the activation-paired chamber (Figures 6E and
6F), but no obvious place preference or aversion was observed
after activating ChAT+ neurons (Figures 6G and 6H). These
results suggest that VgluT2+ neurons in the SI are most likely
to receive aaversive signals from the CeL. This speculation
was corroborated after AAV2/9-FLEX-taCasp3-TEVp was
infused into the SI of VgluT2-Cre mice selectively to kill VgluT2+
neurons and AAV9-CAG-ChR2-GFP was infused into the CeL of
the same mice to activate the CeL-SI circuit (Figure 6I). Consis-
tent with the effects of taCasp3 on inducing cell apoptosis (Yang
et al., 2013), a mass of ablated neurons was observed by TUNEL
apoptosis detection (Figure 6J). SI-VgluT2-taCasp3 mice had
significantly reduced CPA induced by activation of the CeL-SI
circuit (Figures 6K and 6L), suggesting that SIVgluT2+ neurons,
as a relay station receiving aversive signals from CeLPKC-d+ neu-
rons, are indispensable for aversive-related behavior induced by
activation of CeLPKC-d+ neurons.

F(1, 11) = 17.26, **p = 0.002; green line: F(1, 10) = 8.23, *p = 0.017. (G) Black line: F(1, 11) = 18.11, **p = 0.001; green line: F(1, 10) = 9.76, *p = 0.011. Data represent
means ± SEM.
(H) Schematic of targeted genetic inactivation of the CeLPKC-d+-SI circuit.
(I) Representative images of the AAV injection site in the CeL (top) and ArchT-expressing terminals in the SI (bottom). Scale bar, 500 mm.
(J–M) Performance (J), discriminability (K), hit rate (L), and correct rejection rate (M) of ArchT (n = 8) and control mice (n = 8). n.s., not significant (two-way ANOVA
with mixed design). (J) F(1, 14) = 37.90, ***p < 0.001; (K) F(1, 14) = 16.81, **p = 0.001; (M) F(1, 14) = 42.35, ***p < 0.001. Data represent means ± SEM.
(N) Distribution of the response delay in the go trials.
(O) Median response delay of mice.
(P and Q) Performance (P) and correct-rejection rate (Q) of mice in the last session of learning without laser (training) and with laser (expression). n.s., not
significant (Mann-Whitney U test). Boxplots show mean (solid square), median, quartiles (boxes), and range (whiskers).

Cell Reports 21, 1770–1782, November 14, 2017 1777

Figure 6. SIVgluT2+ Neurons Are Indispensable for CPA Produced by CeLPKC-d+ Neurons
(A) Procedure for RTPP performance.
(B) Examples of pre-test and real-time behavior traces accompanying photostimulation of SI VgluT2+ neurons.
(C–H) Aversion or preference for place was shown when SI VgluT2+ (C and D, VgluT2-GFP, n = 4; VgluT2-ChR2, n = 4), GABA+ (E and F, GABA-GFP, n = 5; GABA-
ChR2, n = 4), and ChAT+ neurons (G and H, ChAT-GFP, n = 4; ChAT-ChR2, n = 5) were activated. n.s., not significant (Mann-Whitney U test). VgluT2+: RT/Pre,
*p = 0.021; avoidance score, *p = 0.021. GABA+: RT/Pre, *p = 0.027; avoidance score, *p = 0.014.
(I) Diagram of viruses injected into the SI and CeL. Optical fiber was implanted above the SI for photostimulation of CeLPKC-d+ outputs.

(legend continued on next page)

1778 Cell Reports 21, 1770–1782, November 14, 2017

DISCUSSION
Combining the head-fixed go/no-go behavioral paradigm, fiber
photometry, optogenetic manipulation, and in vivo electrophys-
iology and anatomy, we characterized a CeL-SI neural circuit
that encodes aversive signals and serves as a negative reinforcer
to bi-directionally modulate associative learning. Importantly, we
found that CeLPKC-d+ neurons drive CPA by transmitting an aver-
sive signal to SI glutamatergic neurons.
Prior studies have demonstrated amygdala responses evoked
by aversive stimuli, but the specific neural type has remained
unclear (Calu et al., 2010; McHugh et al., 2014). Neurons ex-
pressing calcitonin gene-related peptide receptor (CGRPR) in
the CeL, which can receive footshock-driven US aversive signals
from parabrachial nucleus CGRP neurons, have been found to
partially overlap with PKC-d+ neurons (Han et al., 2015). Consid-
ering its expression in multiple neural populations, the CGRPR as
a marker to study US aversive signal transmission from the CeL
to other downstream nodes might not be particularly accurate
(Han et al., 2015). Using fiber photometry, we identified a sub-
population of CeL neurons expressing PKC-d that showed a
robust response to the air puff as an aversive stimulus during
negative reinforcement learning (Figure 1). The rapid increase
in GCaMP5g signal activity following punishment indicated
some homogeneity among the response patterns of the
PKC-d+ neurons encoding an aversive signal. At the same
time, the response to aversive stimuli in the CeLPKC-d+ neurons
weakly decreased as training proceeded, leading to the specu-
lation that CeLPKC-d+ neurons might be related to teaching
signals (Johansen and Fields, 2004; McNally et al., 2011). A prev-
alent model for the functional organization of amygdala circuits
posits CeL PKC-d+ neurons as CeLoff neurons, a population
that shows inhibitory CS responses following fear conditioning
(Haubensak et al., 2010). On the contrary, we did not observe in-
hibition of CeL PKC- d+ neural activity in response to CS during
the whole training phases. Weak increase of CeL PKC-d+ neural
activity in response to CS was observed during the early
learning, which might be caused by the possible that mice
mistake that the CS is aversive. It is noteworthy that we found
that a substantial number of PKC-d+ neurons were aversive
‘‘US-on’’ neurons and modulate negative reinforcement learning
by conveying an aversive US, implying that CeLPKC-d+ neurons
might encode different signals in different behavioral tasks.
A weak (nonsignificant) increase in GCaMP5g fluorescence in
the early stage and suppression in the later and expression
stages were also observed when mice licked to acquire a
reward, suggesting that CeL PKC-d+ neurons might participate
in valence-specific behaviors. Although CeL SOM+ neurons
were activated by threat-predicting sensory cues after fear con-
ditioning (Yu et al., 2016), we did not observe significant re-
sponses to the aversive CS (Figure S2J). This discrepancy could
be due to different behavioral tasks. In fact, CeA SOM+ neurons
(J) Images show apoptosis neurons by TUNEL apoptosis detection after AAV-flex-taCasp3-TEVp or AAV2-EF1a-Flex-GFP was injected into the SI of VgluT2-Cre
mice. Scale bar, 20 mm.
(K and L) Preference (K) and time spent (L) in the conditioned chamber after ablating SI VgluT2+ neurons compared with control (VgluT2-GFP, n = 7; VgluT2-
taCasp3, n = 6). RT/Pre, VgluT2-GFP: RT/Pre, *p = 0.022; avoidance score, *p = 0.032 (Mann-Whitney U test). Boxplots show mean (solid square), median,
quartiles (boxes), and range (whiskers).
have recently been shown to fire in different patterns in response
to conditioned stimuli with different behavioral responses (Fadok
et al., 2017). Thus, to gain better understanding of the heteroge-
neity of responses in CeL PKC-d+ and SOM+ neurons, it would
be interesting to know the temporal profile of firing rate changes
by multi-channel recording or imaging individual cells in different
tasks in future research.
Knowledge about the functions of the CeL-SI neural circuit
remains limited. Previous studies have suggested that CeA
efferents and BF neurons mainly overlap and form symmetric
synapses with neurons in the ventrolateral SI (Jolkkonen et al.,
2002). Our anatomy results revealed the contribution of PKC-
d+ neurons to the SI-projecting cell population (Figure 3) and
thus contribute to a more comprehensive understanding of the
CeL-BF circuit. The CeLPKC-d+ neurons mediated negative rein-
forcement learning by relaying an aversive signal to the SI. This
result extends and complements the extensive roles of CeL
neurons in encoding incentive motivation, conditioned reinforce-
ment, and fear conditioning (Ciocchi et al., 2010; Haubensak
et al., 2010; Li et al., 2013; Tye and Janak, 2007; Yu et al.,
2016). We observed that CeL PKC-d+ neurons also projected
to other brain regions, such as the BNSTv, which is a major target
of projections from the BLA and CeA and plays an important role
in mediating negative emotion, such as anxiety (Duvarci et al.,
2009). Inhibition of CeLPKC-d+-BNST circuit activity had no signif-
icant effect on performance in the go/no-go task (data not
shown). Considering that CeL PKC-d+ neurons can mediate
appetitive behavior such as feeding (Cai et al., 2014), our data
raise several questions: What is the underlying neural circuit for
reward-related signals in the amygdala? What are the differ-
ences between circuits transmitting appetitive signals and those
transmitting aversive signals? We will expand our current
research in future studies to map whole-brain monosynaptic
inputs to the amygdala and BF and manipulate those inputs
separately to test their effects on associative learning.
Multi-channel results showed that certain neurons in the SI
that were activated by enhancing CeL-SI circuit activity exhibited
long latency or even double activation (Figure S5), which
suggests that functional connection from the CeLPKC-d+ neurons
to the SI might be executed by multi-synaptic pathways of the SI,
thus highlighting the complexity of SI local circuits. Immuno-
staining revealed that photoactivation of the CeL-SI circuit
induced c-Fos expression in SI VgluT2+ neurons. Rabies-virus-
mediated monosynaptic retrograde tracing suggested that CeL
neurons could target three cell types in SI (Figure 2D). As
CeLPKC-d+ outputs are GABAergic (Haubensak et al., 2010), the
most parsimonious explanation consistent with these results in-
volves a local SI disinhibitory circuitry mechanism that could
convert an inhibitory input from PKC-d+ neurons to activate SI
VgluT2+ neurons. Besides, by selectively ablating VgluT2+ neu-
rons in the SI, we revealed that these neurons are essential for
CeLPKC-d+-SI neural circuitry in driving CPA. Taken together,
Cell Reports 21, 1770–1782, November 14, 2017 1779
these data suggest that SI VgluT2+ neurons are responsible for
receiving an aversive signal from CeL PKC-d+ neurons. We
could not exclude the possibility that the SI VgluT2+ neurons
can also be activated by behavior (CPA) elicited by CeL PKC-
d+ neuron activation or that other SI neural subtypes also can
receive aversive signals from CeL PKC-d+ neurons but are
unable to drive a behavioral response. Previous studies have
mainly focused on the functions of the cholinergic system
involved in cortical plasticity, spatial learning, working memory,
and other mnemonic tasks (Chudasama et al., 2004; Do et al.,
2016; Kilgard, 2003). Studies on the functions of BF non-cholin-
ergic neurons forming widespread inputs and projections are
limited (Do et al., 2016). Our optogenetic manipulation revealed
that activating SI glutamatergic neurons elicited obvious CPA,
whereas activating SI GABA+ neurons elicited CPP, suggesting
that both cell types might perform contrary functions in some
contexts. In summary, our study of the neural circuitry involved
in transmitting aversive signals furthers our current understand-
ing of the functions of the amygdala in encoding aversive asso-
ciative learning.
EXPERIMENTAL PROCEDURES
Animals
Male and female mice 4–12 weeks of age were used for the experiments. All
mice used for go/no-go behavioral training were bred onto the C57BL/6J
genetic background. The Som-IRES-Cre, VgluT2-Cre, and GAD2-Cre mice
were obtained from the Josh Huang Lab (Cold Spring Harbor Laboratory,
Cold Spring Harbor, NY, USA). Rosa26-loxP-STOP-loxP-H2B-EYFP reporter
line (Ai3) and ChAT-IRES-Cre mice (Jackson Laboratories stock number
006410) were purchased from The Jackson Laboratory. PKC-d-Cre mice
were provided from the Bo Li Lab (Cold Spring Harbor Laboratory). Som-
IRES-Cre mice were crossed with Ai3 mice to visualize SOM+ neurons. All
mice were group-housed under a 12-hr light/dark cycle (7:00 a.m. to 7:00
p.m. light) and bred in the animal facility at the Wuhan National Laboratory
for Optoelectronics (Wuhan, China). All procedures involving animals were
approved by the Hubei Provincial Animal Care and Use Committee and
complied with the experimental guidelines of the Animal Experimentation
Ethics Committee of Huazhong University of Science and Technology, China.
Virus and Retrograde Tracer
The retrograde tracer Alexa Fluor 488/647-conjugated CTB was purchased
from Invitrogen. AAV viruses, including AAV9-EF1a-Flex-ArchT-GFP,
AAV9-CAG-ChR2-GFP, AAV2-EF1a-DIO-EYFP, AAV2-EF1a-Flex-GFP, and
AAV5-EF1a-DIO-hChR2-EYFP, were produced at the Gene Therapy Center
Vector Core Facility, University of North Carolina (Chapel Hill, NC, USA).
AAV2-FLEX-tasCasp3-TEVp was purchased from Shanghai Taiting Biolog-
ical (Shanghai, China). AAV2-EF1a-DIO-GCaMP5g was purchased from
Obio Technology (Shanghai, China). The monosynaptic retrograde tracing
system, including AAV5-CAG-FLEX TVA66T-mCherry, AAV5-CAG-FLEX-
RG, and rabies virus EnvA-RV-YFP, was provided by the Wuhan Institute
of Physics and Mathematics (WIPM), Chinese Academy of Sciences (CAS).
For detailed viral injection procedures, see Supplemental Experimental
Procedures.
The Go/No-Go Behavioral Task
The go/no-go task included a shaping phase and a training phase. Mice were
trained for two sessions per day. Each session consisted of 200 trials. Air puff,
water delivery, cue, and laser illumination were triggered through serial ports,
and licking events were recorded by a recording system (LabState ver1.0 for
operant behavior, AniLab Software & Instruments, China). For details on calcu-
lation of performance, the optical stimulation protocol, and additional informa-
tion, see Supplemental Experimental Procedures.
1780 Cell Reports 21, 1770–1782, November 14, 2017
Immunohistochemistry and Cell Quantification
Immunohistochemical analysis was performed as described previously (Li et al.,
2011). Primary antibodies were anti-PKC-d (mouse, BD Biosciences, 610397,
1:700) and anti-c-Fos (rabbit, Synaptic Systems, 226003, 1:10,000). Detailed
experimental methods can be found in Supplemental Experimental Procedures.
Fiber Photometry
Following AAV2-EF1a-DIO-GCaMP5g or AAV2-EF1a-Flex-GFP virus injec-
tion, an optical fiber (230 mm outer diameter [O.D.], 0.37 numerical aperture
[NA]; Shanghai Fiblaser Technology, Shanghai, China) was placed in a ceramic
ferrule and inserted toward the CeL. The ceramic ferrule was supported with
dental acrylic. Mice were allowed to recover for at least 1 week. For recording
fluorescence signals, we used the fscope fiber photometry system
(BiolinkOptics, China). An optical fiber (230 mm O.D., NA 0.37; Thorlabs) guided
the 473-nm laser between the recording system and the implanted optical
fiber. The laser power was adjusted at 20–30 mW. GCaMP5g fluorescence
was further filtered through a low-pass filter (30 Hz cutoff). Signals were
collected at a sampling frequency of 100 Hz. The relative change in fluores-
cence, DF/F = (F F0)/F0, was calculated by taking F0 to be the baseline of
the fluorescence signal averaged over 4 s before CS (light or tone) in each ses-
sion. For analysis of the statistical significance of event-related fluorescence
changes, we defined the mean of the signals during the 0.5 s period before
CS onset as the baseline fluorescence signal and the US (air puff and water)
evoked responses as the mean of the fluorescence signals for the duration
of the US. Considering the peak US responses with delays longer than the
duration of US, we searched the peak US responses in each session within
a 1-s time window after US onset. All data were recorded by f-scopeACQ
software and analyzed using MATLAB.
CPA
CPA was performed in a two-compartment conditioning apparatus (25 3 25 3
50 cm/chamber). A video-tracking system (ANY-Maze software, Stoelting,
Wood Dale, IL, USA) recorded all mice movements. For detailed behavioral
procedures, see Supplemental Experimental Procedures.
Statistics
Sample Size
We did not perform a calculation on sample size. Sample size was based on
past experience using similar techniques and animal models.
Analysis
Data are presented as means ± SEM or as box-and-whisker plots. Data collec-
tion and analysis were not performed blindly. We did not randomize subjects.
They were assigned to control or experimental groups based on whether they
received EYFP, GFP, ChR2, or ArchT virus injection. p values were calculated
by Mann-Whitney U test. For comparisons across two learning curves, data
were analyzed using two-way analysis of variance with mixed design. All
data were analyzed using SPSS software (version 22, IBM, New York, NY,
USA). Statistical significance was set at *p < 0.05, **p < 0.01, and ***p < 0.001.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures
and six figures and can be found with this article online at https://doi.org/10.
1016/j.celrep.2017.10.062.
AUTHOR CONTRIBUTIONS
Y.C., G.L., and S.J. performed most of the experiments. Y.C., G.L., and H.L.
analyzed the data. J.Y., J.P., X.H., and H.G. performed partial imaging exper-
iments. H.L., F.X., and T.X supervised the project. Y.C. and H.L. designed the
study and wrote the manuscript.
ACKNOWLEDGMENTS
We thank B. Li (Cold Spring Harbor Lab) for the PKC-d-Cre mice, Y.Y. Li (Huaz-
hong University of Science & Technology) for help with the MATLAB coding,
and L.W. Mu and Q.L. Hu (Huazhong University of Science & Technology) for
optogenetic setups. We thank Kristen Delevich (UC Berkeley) and Xiaohong Xu
(Institute of Neuroscience, CAS) for critical reading and editing of the manu-
script, Hailan Hu (Zhejiang University) for critical reading of the manuscript,
and members of the Li laboratory for helpful discussions. This study was sup-
ported by The National Key Research and Development Program of China—
Stem Cell and Translational Research (2016YFA0102500), the National Natural
Science Foundation of China (grants 91432107 and 31671105), the Science
Fund for Creative Research Group of China (grant 61421064), and the Director
Fund of the Wuhan National Laboratory for Optoelectronics.
Received: July 12, 2017
Revised: September 26, 2017
Accepted: October 16, 2017
Published: November 14, 2017
REFERENCES
Akerboom, J., Chen, T.W., Wardill, T.J., Tian, L., Marvin, J.S., Mutlu, S., Cal-
derón, N.C., Esposti, F., Borghuis, B.G., Sun, X.R., et al. (2012). Optimization
of a GCaMP calcium indicator for neural activity imaging. J. Neurosci. 32,
13819–13840.
Avarguès-Weber, A., de Brito Sanchez, M.G., Giurfa, M., and Dyer, A.G.
(2010). Aversive reinforcement improves visual discrimination learning in
free-flying honeybees. PLoS ONE 5, e15370.
Avila, I., and Lin, S.C. (2014). Motivational salience signal in the basal forebrain
is coupled with faster and more precise decision speed. PLoS Biol. 12,
e1001811.
Cai, H., Haubensak, W., Anthony, T.E., and Anderson, D.J. (2014). Central
amygdala PKC-d(+) neurons mediate the influence of multiple anorexigenic
signals. Nat. Neurosci. 17, 1240–1248.
Calu, D.J., Roesch, M.R., Haney, R.Z., Holland, P.C., and Schoenbaum, G.
(2010). Neural correlates of variations in event processing during learning in
central nucleus of amygdala. Neuron 68, 991–1001.
Chen, T.W., Wardill, T.J., Sun, Y., Pulver, S.R., Renninger, S.L., Baohan, A.,
Schreiter, E.R., Kerr, R.A., Orger, M.B., Jayaraman, V., et al. (2013). Ultrasen-
sitive fluorescent proteins for imaging neuronal activity. Nature 499, 295–300.
Chudasama, Y., Dalley, J.W., Nathwani, F., Bouger, P., and Robbins, T.W.
(2004). Cholinergic modulation of visual attention and working memory: disso-
ciable effects of basal forebrain 192-IgG-saporin lesions and intraprefrontal in-
fusions of scopolamine. Learn. Mem. 11, 78–86.
Ciocchi, S., Herry, C., Grenier, F., Wolff, S.B., Letzkus, J.J., Vlachos, I., Ehrlich,
I., Sprengel, R., Deisseroth, K., Stadler, M.B., et al. (2010). Encoding of condi-
tioned fear in central amygdala inhibitory circuits. Nature 468, 277–282.
Ciocchi, S., Passecker, J., Malagon-Vina, H., Mikus, N., and Klausberger, T.
(2015). Brain computation. Selective information routing by ventral hippocam-
pal CA1 projection neurons. Science 348, 560–563.
Cybulska-Klosowicz, A., Zakrzewska, R., and Kossut, M. (2009). Brain activa-
tion patterns during classical conditioning with appetitive or aversive UCS.
Behav. Brain Res. 204, 102–111.
Davis, M., Walker, D.L., and Myers, K.M. (2003). Role of the amygdala in fear
extinction measured with potentiated startle. Ann. N Y Acad. Sci. 985,
218–232.
Do, J.P., Xu, M., Lee, S.H., Chang, W.C., Zhang, S., Chung, S., Yung, T.J., Fan,
J.L., Miyamichi, K., Luo, L., et al. (2016). Cell type-specific long-range connec-
tions of basal forebrain circuit. eLife 5, e13214.
Dolzani, S.D., Nakamura, S., and Cooper, D.C. (2013). A novel variable delay
Go/No-Go task to study attention, motivation and working memory in the
head-fixed rodent. F1000Res 2, 125.
Duvarci, S., Bauer, E.P., and Paré, D. (2009). The bed nucleus of the stria ter-
minalis mediates inter-individual variations in anxiety and fear. J. Neurosci. 29,
10357–10361.
€thi, A. (2009).
Ehrlich, I., Humeau, Y., Grenier, F., Ciocchi, S., Herry, C., and Lu
Amygdala inhibitory circuits and the control of fear memory. Neuron 62,
757–771.
Fadok, J.P., Krabbe, S., Markovic, M., Courtin, J., Xu, C., Massi, L., Botta, P.,
Bylund, K., Mu €ller, C., Kovacevic, A., et al. (2017). A competitive inhibitory cir-
cuit for selection of active and passive fear responses. Nature 542, 96–100.
Fanselow, M.S., and Poulos, A.M. (2005). The neuroscience of mammalian
associative learning. Annu. Rev. Psychol. 56, 207–234.
Gomez, P., Ratcliff, R., and Perea, M. (2007). A model of the go/no-go task.
J. Exp. Psychol. Gen. 136, 389–413.
Gozzi, A., Jain, A., Giovannelli, A., Bertollini, C., Crestan, V., Schwarz, A.J.,
Tsetsenis, T., Ragozzino, D., Gross, C.T., and Bifone, A. (2010). A neural switch
for active and passive fear. Neuron 67, 656–666.
Han, S., Soleiman, M.T., Soden, M.E., Zweifel, L.S., and Palmiter, R.D. (2015).
Elucidating an Affective Pain Circuit that Creates a Threat Memory. Cell 162,
363–374.
Hangya, B., Ranade, S.P., Lorenc, M., and Kepecs, A. (2015). Central cholin-
ergic neurons are rapidly recruited by reinforcement feedback. Cell 162, 1155–
1168.
Harrison, T.C., Pinto, L., Brock, J.R., and Dan, Y. (2016). Calcium imaging of
basal forebrain activity during innate and learned behaviors. Front. Neural Cir-
cuits 10, 36.
Haubensak, W., Kunwar, P.S., Cai, H., Ciocchi, S., Wall, N.R., Ponnusamy, R.,
Biag, J., Dong, H.W., Deisseroth, K., Callaway, E.M., et al. (2010). Genetic
dissection of an amygdala microcircuit that gates conditioned fear. Nature
468, 270–276.
Henny, P., and Jones, B.E. (2008). Projections from basal forebrain to prefron-
tal cortex comprise cholinergic, GABAergic and glutamatergic inputs to pyra-
midal cells or interneurons. Eur. J. Neurosci. 27, 654–670.
Herry, C., Ciocchi, S., Senn, V., Demmou, L., Mu €ller, C., and Lu
€thi, A. (2008).
Switching on and off fear by distinct neuronal circuits. Nature 454, 600–606.
Johansen, J.P., and Fields, H.L. (2004). Glutamatergic activation of anterior
cingulate cortex produces an aversive teaching signal. Nat. Neurosci. 7,
398–403.
Jolkkonen, E., Miettinen, R., Pikkarainen, M., and Pitkänen, A. (2002). Projec-
tions from the amygdaloid complex to the magnocellular cholinergic basal
forebrain in rat. Neuroscience 111, 133–149.
Kilgard, M. (2003). Cholinergic modulation of skill learning and plasticity.
Neuron 38, 678–680.
Lau, B., and Salzman, C.D. (2008). Noncholinergic neurons in the basal fore-
brain: often neglected but motivationally salient. Neuron 59, 6–8.
LeDoux, J.E. (2000). Emotion circuits in the brain. Annu. Rev. Neurosci. 23,
155–184.
Li, A., Gong, H., Zhang, B., Wang, Q., Yan, C., Wu, J., Liu, Q., Zeng, S., and
Luo, Q. (2010). Micro-optical sectioning tomography to obtain a high-resolu-
tion atlas of the mouse brain. Science 330, 1404–1408.
Li, B., Piriz, J., Mirrione, M., Chung, C., Proulx, C.D., Schulz, D., Henn, F., and
Malinow, R. (2011). Synaptic potentiation onto habenula neurons in the learned
helplessness model of depression. Nature 470, 535–539.
Li, H., Penzo, M.A., Taniguchi, H., Kopec, C.D., Huang, Z.J., and Li, B. (2013).
Experience-dependent modification of a central amygdala fear circuit. Nat.
Neurosci. 16, 332–339.
Lin, S.C., and Nicolelis, M.A.L. (2008). Neuronal ensemble bursting in the basal
forebrain encodes salience irrespective of valence. Neuron 59, 138–149.
Mayo, W., Dubois, B., Ploska, A., Javoy-Agid, F., Agid, Y., Le Moal, M., and
Simon, H. (1984). Cortical cholinergic projections from the basal forebrain of
the rat, with special reference to the prefrontal cortex innervation. Neurosci.
Lett. 47, 149–154.
~o, L., Lima, J., Lowry, J.P., and
McHugh, S.B., Barkus, C., Huber, A., Capita
Bannerman, D.M. (2014). Aversive prediction error signals in the amygdala.
J. Neurosci. 34, 9024–9033.
Cell Reports 21, 1770–1782, November 14, 2017 1781
McNally, G.P., Johansen, J.P., and Blair, H.T. (2011). Placing prediction into
the fear circuit. Trends Neurosci. 34, 283–292.
Miyamichi, K., Amat, F., Moussavi, F., Wang, C., Wickersham, I., Wall, N.R.,
Taniguchi, H., Tasic, B., Huang, Z.J., He, Z., et al. (2011). Cortical representa-
tions of olfactory input by trans-synaptic tracing. Nature 472, 191–196.
O’Doherty, J., Dayan, P., Schultz, J., Deichmann, R., Friston, K., and Dolan,
R.J. (2004). Dissociable roles of ventral and dorsal striatum in instrumental
conditioning. Science 304, 452–454.
Pape, H.C. (2010). Petrified or aroused with fear: the central amygdala takes
the lead. Neuron 67, 527–529.
Paré, D., Quirk, G.J., and Ledoux, J.E. (2004). New vistas on amygdala net-
works in conditioned fear. J. Neurophysiol. 92, 1–9.
Peck, C.J., and Salzman, C.D. (2014). The amygdala and basal forebrain as a
pathway for motivationally guided attention. J. Neurosci. 34, 13757–13767.
Resnik, J., Sobel, N., and Paz, R. (2012). Auditory aversive learning increases
discrimination thresholds. J. Mol. Neurosci. 48, S96–S96.
Tye, K.M., and Janak, P.H. (2007). Amygdala neurons differentially encode
motivation and reinforcement. J. Neurosci. 27, 3937–3945.
1782 Cell Reports 21, 1770–1782, November 14, 2017
Wall, N.R., Wickersham, I.R., Cetin, A., De La Parra, M., and Callaway, E.M.
(2010). Monosynaptic circuit tracing in vivo through Cre-dependent targeting
and complementation of modified rabies virus. Proc. Natl. Acad. Sci. USA
107, 21848–21853.
Wickersham, I.R., Finke, S., Conzelmann, K.K., and Callaway, E.M. (2007).
Retrograde neuronal tracing with a deletion-mutant rabies virus. Nat. Methods
4, 47–49.
Xu, D., Jiang, T., Li, A., Hu, B., Feng, Z., Gong, H., Zeng, S., and Luo, Q. (2013).
Fast optical sectioning obtained by structured illumination microscopy using a
digital mirror device. J. Biomed. Opt. 18, 060503.
Yang, C.F., Chiang, M.C., Gray, D.C., Prabhakaran, M., Alvarado, M., Juntti,
S.A., Unger, E.K., Wells, J.A., and Shah, N.M. (2013). Sexually dimorphic neu-
rons in the ventromedial hypothalamus govern mating in both sexes and
aggression in males. Cell 153, 896–909.
Yu, K., Garcia da Silva, P., Albeanu, D.F., and Li, B. (2016). Central amygdala
somatostatin neurons gate passive and active defensive behaviors.
J. Neurosci. 36, 6488–6496.