LETTER
1038/nature21047
A competitive inhibitory circuit for selection of
active and passive fear responses
Jonathan P. Fadok1, Sabine Krabbe1, Milica Markovic1,2, Julien Courtin1, Chun Xu1, Lema Massi1, Paolo Botta1,2†, Kristine Bylund1,
Christian Müller1, Aleksandar Kovacevic1, Philip Tovote1 & Andreas Lüthi1,2
When faced with threat, the survival of an organism is contingent
upon the selection of appropriate active or passive behavioural
responses1–3. Freezing is an evolutionarily conserved passive fear
response that has been used extensively to study the neuronal
mechanisms of fear and fear conditioning in rodents4. However,
rodents also exhibit active responses such as flight under natural
conditions2. The central amygdala (CEA) is a forebrain structure
vital for the acquisition and expression of conditioned fear
responses, and the role of specific neuronal sub-populations of
the CEA in freezing behaviour is well-established1,5–7. Whether
the CEA is also involved in flight behaviour, and how neuronal
circuits for active and passive fear behaviour interact within the
CEA, are not yet understood. Here, using in vivo optogenetics and
extracellular recordings of identified cell types in a behavioural
model in which mice switch between conditioned freezing and
flight, we show that active and passive fear responses are mediated
by distinct and mutually inhibitory CEA neurons. Cells expressing
corticotropin-releasing factor (CRF+) mediate conditioned flight,
and activation of somatostatin-positive (SOM+) neurons initiates
passive freezing behaviour. Moreover, we find that the balance
between conditioned flight and freezing behaviour is regulated by
means of local inhibitory connections between CRF+ and SOM+
neurons, indicating that the selection of appropriate behavioural
responses to threat is based on competitive interactions between two
defined populations of inhibitory neurons, a circuit motif allowing
for rapid and flexible action selection.
Mammals express adaptive behavioural responses to threat1,2. The
type of response depends on intensity and proximity of threat, as
well as context2,8. Although different behavioural fear responses are
inborn, associative learning enables organisms to use environmental
cues to predict and respond to danger9. Pavlovian fear conditioning is
one ­paradigm that has been used to understand the neural circuitry
underlying associative learning5,9,10. In Pavlovian fear conditioning, a
conditioned stimulus (CS; for example, an auditory stimulus) is paired
with an aversive unconditioned stimulus (US; for example, electrical
footshock). Electrical footshock, the most commonly used US, elicits a
variety of defensive responses, including flight and freezing4; ­however,
freezing is the dominant conditioned response to the CS in most
fear-conditioning experiments5.
We developed a Pavlovian conditioning paradigm that promotes
active defensive behaviour, including flight. Conditioned flight
responses to the CS were elicited by extending standard conditioning
protocols (Extended Data Fig. 1). To allow for direct, within-subject
comparison of active and passive behaviours, we tested different CSs
and found that a serial compound stimulus (SCS) consisting of a pure
tone followed by white noise induced rapid behavioural switches
between conditioned freezing and flight behaviour (Fig. 1a–d). With
conditioning, mice exhibited freezing in response to the tone, and
flight during the white noise (Fig. 1c, d, Supplementary Video 1). The
1
Friedrich Miescher Institute for Biomedical Research, Maulbeerstrasse 66, 4058 Basel, Switzerland. 2University of Basel, 4000 Basel, Switzerland. †Present address: Champalimaud Centre for the
Unknown, Av. Brasilia, 1400-038 Lisbon, Portugal.
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
doi:10.
conditioned flight response was characterized by increases in speed
(Fig. 1d) and the number of escape jumps (Fig. 1d, Supplementary
Video 1). This behaviour was quantified as a flight score (Fig. 1c, d, see
Methods). Behavioural analysis revealed that tone exposure signifi-
cantly enhanced freezing compared to contextual freezing, and that
exposure to white noise, concomitant with the observed increase in
flight score, resulted in a significant reduction of the time spent freezing
(Fig. 1d). Importantly, the white noise stimulus was not in itself aversive
(Extended Data Fig. 1).
To test for the effect of threat proximity and context2,8 on conditioned
flight, mice were split into two groups: one that underwent extinction
training in the conditioning context, and one that was exposed to the
SCS in a neutral context (Fig. 1a, e, f). Consistent with threat immi-
nence theory8, conditioned flight behaviour was rapidly extinguished
upon removal of the footshock during extinction (Fig. 1e), and was also
context-dependent (Fig. 1e, f). In line with the hierarchical organiza-
tion of distinct defensive behaviours2, extinction of conditioned flight
behaviour was accompanied by an increase in freezing in response to
the white noise (Fig. 1e).
The CEA is a striatum-like structure comprised of distinct neuronal
populations of inhibitory GABAergic neurons that express CRF, SOM
or protein kinase C δ​ (PKCδ​)11–14, among other molecular markers15
(see also Extended Data Fig. 2). The CEA can be subdivided into the
lateral CEA (CEl) and medial CEA (CEm). Disinhibition of CEm
output neurons by inhibitory interactions between CEl SOM+ and
PKCδ​+ neurons gates the expression of conditioned freezing11,14,16,17.
Furthermore, suppression of activity within CEA cellular substrates of
freezing leads to active behavioural coping with fear18. To d ­ etermine
the role of defined CEA neuronal subpopulations for expression of
conditioned flight, we used the inhibitory opsin a­ rchaerhodopsin
(Arch), which was Cre-dependently expressed in CRF+, SOM+
and PKCδ​+ neurons (Fig. 2a and Extended Data Fig. 2). Efficacy of
Arch-mediated neuronal inhibition was tested using extracellular
recordings of identified units in behaving animals (Extended Data
Fig. 2, see also Methods). Inhibition of CRF+, but not SOM+ or
PKCδ​+ neurons, completely abolished conditioned flight behaviour
(Fig. 2b–g and Extended Data Fig. 2). Optogenetic inhibition in naive
mice did not elicit any freezing or flight behaviour (Extended Data Fig. 2).
Fluorophore-expressing somata were predominately found in CEl
(89 ±​ 3% CRF–Arch, 91 ±​ 4% SOM–Arch, and 100% PKCδ​–Arch11);
however, minor expression was also observed in the CEm. Together,
these data demonstrate that CEA CRF+ neurons are necessary for the
expression of conditioned flight.
We next determined the electrophysiological response profile of
CRF+ neurons during conditioned flight. Channelrhodopsin-2 (ChR2)
or Arch was expressed in CEA CRF+ neurons, and mice were subse-
quently implanted with optrodes (Fig. 3a and Extended Data Fig. 3).
Single units were recorded and isolated using standard procedures16
and CRF+ cells were identified by short-latency responses to light
0 0 M O N T H 2 0 1 7 | VO L 0 0 0 | NAT U R E | 1
 RESEARCH LETTER

a Behavioural paradigm Context A

Day 4
b Stimuli (Fig. 3b, Extended Data Fig. 3 and Methods). CRF+ cells exhibited an
Context B
Extinction SCS US increase in responses to white noise between days 1 and 3 (Fig. 3c, d and
Day 1 Day 2 Day 3
16 × SCS
Extended Data Fig. 3); however, in the neutral context, where flight was
4 × SCS 5 × SCS–US 5 × SCS–US
4 × SCS
never observed, CRF+ excitation in response to white noise was dimi­
Pre-exposure Conditioning nished compared to day 3 (Fig. 3e). By normalizing firing rates during
Retrieval Tone White noise
epochs of conditioned flight and freezing behaviour, more than half
c Pre-
exposure Conditioning
d Day 3, conditioning of the CRF+ cells (5 of 8) were found to be excited specifically during
Day 1 Day 2 Day 3
20 flight, but not during freezing behaviour (Fig. 3i and Methods). These
Speed (cm s–1) Pre-CS
18 15
Tone
findings are consistent with the notion that activity of CRF+ neurons
Flight score (AU)

14
10 White noise contributes to conditioned flight behaviour.
10
5 Shock
Previous studies suggested that activation of CEl SOM+ cells is
6
0
0 10 20 30 40
important for conditioned freezing14. We therefore recorded from
2 Time (s) identified CEl SOM+ neurons during the conditioned flight ­paradigm
Pre-CS Tone White noise Shock (Fig. 3a and Extended Data Fig. 3). Before conditioning, there was
Tone
White noise
*** no significant response of identified SOM+ cells to either the pure
100 20
***
20 100 *** ***
*** tone or to white noise (Fig. 3f and Extended Data Fig. 3). Following
Number of jumps

Flight score (AU)

80 80
Freezing (%)

15 15
­conditioning, SOM+ cells became strongly inhibited by white noise
Freezing (%)
60 * 60
40
10 10 (Fig. 3g). However, they exhibited significantly elevated firing rates
40
20 5 5
during the 3 min baseline, a period of high levels of contextual ­freezing
20
0
(pre-­exposure firing rate: 2.7 ±​ 0.6 spikes per s; day 3 firing rate:
Trial 1 2 3 4 1 3 5 1 3 5
0 0 0
5.8 ±​ 1.3 spikes per s; two-tailed paired t-test, P <​  0.05). Consistent
e Day 4, extinction f Day 4, retrieval
with this, SOM+ neurons were excited by both conditioned stimuli
8 20 *** 8 8 in the neutral context, in which mice displayed freezing behaviour in
Flight score (AU)
Flight score (AU)

Flight score (AU)

6 15 6 6 response to both pure tone and white noise (Fig. 3h and Extended Data
4 10 4 4 Fig. 3). When analysing the activity of SOM+ neurons as a function
2 5 2 2 ** of behaviour, we found that the majority of SOM+ neurons (6 of 7)
0 0 0 0 were activated during freezing, but inhibited during flight, pointing
Tone
*** *** towards an interaction between freezing and flight pathways (Fig. 3i
100 White noise 100 100 100
and Methods).
80 80 80 80
To address this, we performed optogenetic gain-of-function experi­
Freezing (%)

Freezing (%)

Freezing (%)

60 60
40 40
20 20
0 0
Trial 1 4 8 12 16
Figure 1 | Conditioned flight paradigm. a, Protocol used to elicit
conditioned flight responses. On day 4, mice were split into two groups.
b, The conditioned stimulus is a serial compound stimulus (SCS) which
is paired with a 1 s footshock during conditioning. c, Top, comparison of
flight scores across sessions (day 1 to day 3, n =​ 20, two-way repeated-
measures ANOVA, cue ×​trial interaction, F(13,247) =​  4.82, P <​  0.01).
AU, arbitrary units. Bottom, comparison of freezing responses across
sessions (day 1 to day 3, n =​ 20; two-way repeated-measures ANOVA,
cue ×​trial interaction, F(13,247) =​  18.3, P <​  0.01). d, Top, average speed
trace of mice (n =​ 10) from day 3, note increase in speed during exposure
to white noise. Bottom left, the number of jump escape responses on day
3 (n =​ 20, Friedman test, P <​ 0.01, Dunn’s multiple comparisons test).
Bottom centre, flight scores on day 3 (n =​ 20, Wilcoxon matched-pairs
signed-rank test). Bottom right, freezing behaviour on day 3 (n =​  20;
one-way repeated-measures ANOVA, F =​  56.82, P <​ 0.01, Tukey’s multiple
comparisons test). e, Top left, rapid extinction of conditioned flight
(n =​ 12, two-way repeated-measures ANOVA, cue ×​trial interaction,
F(15,165) =​  3.05, P <​ 0.01). Top right, flight scores from first block of
four trials of extinction (n =​ 12, Wilcoxon matched-pairs signed-rank
test). Bottom left, freezing in response to white noise increases during
extinction (n =​ 12, two-way repeated-measures ANOVA, cue ×​ trial
interaction, F(15,165) =​  3.55, P <​ 0.01). Bottom right, freezing for the first
block of four trials during extinction (n =​ 12, Wilcoxon matched-pairs
signed-rank test). f, Top left, conditioned flight is context-dependent
(n =​ 8, two-way repeated-measures ANOVA, cue ×​trial interaction,
F(1,7) =​  27.44, P <​ 0.01). Top right, flight scores in the neutral context
(two-tailed paired t-test). Bottom left, freezing responses across trials
during retrieval (n =​ 8, two-way repeated-measures ANOVA, effect of
cue F(1,7) =​  27.67, P <​ 0.01). Bottom right, freezing responses during
retrieval are highest in response to white noise (n =​ 8, two-tailed paired
t-test). Box and whisker plots indicate median, interquartile range,
and 5th to 95th percentiles of the distribution, crosses indicate
means. All other values are means ±​  s.e.m. *​P <​  0.05; *​*​P <​  0.01;
*​*​*​P <​  0.001.
60 60
ments (Fig. 4a and Extended Data Fig. 4). Photo-excitation of ChR2-
40 40
expressing CRF+ neurons during tone exposure decreased freezing and
20 20
led to more active behaviour in the conditioning context (Fig. 4b–d),
0
Trial 1–4 Trial 1 2 3 4
0
Trial 1–4 and decreased freezing in response to the SCS in the neutral context
(Extended Data Fig. 4). There was no effect on freezing or flight in naive
animals (Extended Data Fig. 4). Therefore, in conditioned mice, exo­
genous excitation of CRF+ neurons is sufficient to bias fear responses
away from passive freezing behaviour and towards more active defen-
sive behaviour.
Conversely, optical excitation of SOM+ neurons during exposure to
white noise decreased flight and caused an increase in freezing behaviour
(Fig. 4b–d), in line with the observed neuronal activation patterns (Fig. 3),
and a recent finding suggesting a role for this neuronal subtype in
modu­lating behavioural output19. Consistent with previous reports14,
we also found that ChR2-mediated stimulation of SOM+ cells induced
unconditioned freezing in naive SOM–ChR2 mice (Extended Data
Fig. 4).
One possible mechanism that could account for the mutual exclusion
between the freezing and the flight behaviours could be a reciprocal
inhibitory connection between CRF+ and SOM+ neurons within the
local CEA network. To test this, we expressed ChR2–eYFP in CEA
CRF+ cells and recorded inhibitory postsynaptic currents (IPSCs) in
non-labelled CEl target neurons of acute brain slices during light stimu-
lation of the CRF+ network (Fig. 4e–h and Extended Data Fig. 5). While
both SOM+ and PKCδ​+ neurons receive GABAergic inhibitory input
from CRF+ cells (Fig. 4f, h and Extended Data Fig. 5), we observed
that there were significantly more SOM+ than PKCδ​+ cells receiving
inhibition from CRF+ neurons (Fig. 4g). These data suggest that the
CRF+ network, which is vital for conditioned flight, provides direct
inhibition to the SOM+ network, which encodes freezing behaviour.
We next optically activated SOM+ neurons and recorded from
identified CRF+ and non-labelled target neurons in the CEl (Fig. 4i–l
and Extended Data Fig. 5). These experiments revealed predominant
GABAergic inhibitory connections from SOM+ onto CRF+ neurons
(Fig. 4j–l and Extended Data Fig. 5). Importantly, evoked spiking
activity of both CRF+ and SOM+ neurons could be blocked by

2 | NAT U R E | VO L 0 0 0 | 0 0 M O N T H 2 0 1 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 LETTER RESEARCH

a AAV–flex–Arch–GFP or b Behavioural paradigm

a AAV–flex–Arch–GFP or b
AAV–flex–GFP
AAV flex GFP Fibre AAV–flex–ChR2–eYFP Fibre Electrode 3
Day 1 Day 2 Day 3

Jitter (ms)
connector connector connector
2
4 × SCS 5 × SCS–US 8 × SCS–US
Spont. 1
Optrode 50 μV Light evok.
Pre-exposure Conditioning 250 μs 0

Waveform similarity (r)

Light CRF SOM
Day 3 10 s 10 s
1.000

Frequency (Hz)
4 × SCS
60
0.975
BLA Mouse lines: Light on BLA 30
CEA CRF-ires-cre CEA Mouse lines:
4 × SCS 0 0.950
SOM-ires-cre White noise CRF-ires-cre 0 0.3
White noise + light SOM-ires-cre Time (s)
c d Day 3 onset g Day 3 white noise
c d e

Change in flight score (%)

8 100 Pre-exposure (day 1) Conditioning (day 3) Retrieval (day 4)
CRF–Arch ** White noise White noise White noise
CEl White noise
Speed (cm s–1)

6 White noise 50 **
+ light
4 0
BLA
2 –50 4
CEm CRF+
3 cells

z score
0 –100 2
GFP control
e f Day 3 h CRF–Arch
1
8 100
Change in freezing (%)

SOM–Arch SOM–Arch 0
CEl
White noise –1
Speed (cm s–1)

6 50
White noise
+ light f g h
4 0
BLA
CEm 2 –50
4 SOM+
0 –100
cells

z score
–10
10 –5
5 0 5 10 2
Time (s)
0
Figure 2 | CRF+ neurons mediate conditioned flight. a, Injection and –2
implantation strategy for loss-of-function experiments. Mouse brain figure –250 0 250 500 –250 0 250 500 –250 0 250 500
reproduced with permission from ref. 31. BLA, basolateral amygdala. Time (ms) Time (ms) Time (ms)
b, Behavioural protocol design for loss-of-function experiments.
i Flight Freezing Flight Freezing
c, Expression of Arch in CRF+ neurons. Bregma −​1.58 mm caudal. Scale 4
bar, 250 μ​m. d, Speed traces of the CRF–Arch group (n =​  7). e, Expression 2
of Arch in SOM+ neurons. Bregma −​1.58 mm caudal. Scale bar, 250 μ​m.
z score

f, Speed traces of the SOM–Arch group (n =​  7). g, Flight scores are lowered 0
by optical inhibition of CRF+, but not SOM+ cells. (GFP control, n =​  6; –2
CRF, n =​  7; SOM, n =​ 7; one-way ANOVA, F =​  9.52, P <​  0.01. Tukey’s CRF+ cells SOM+ cells
multiple comparisons test). h, Change in freezing induced by optical 0 5 10 15 0 5 10 15 0 5 10 15 0 5 10 15
inhibition of CRF+ or SOM+ cells (one-way ANOVA; no significant Time bins Time bins Time bins Time bins
differences). Box and whisker plots indicate median, interquartile range, Figure 3 | Neuronal activity patterns of CRF+ and SOM+ cells.
and 5th to 95th percentiles of the distribution, crosses indicate means. All a, Injection and implantation strategy for identified single-unit recordings.
other values are means ±​  s.e.m. *​*​P <​  0.01. Mouse brain figure reproduced with permission from ref. 31. b, Top left,
example spike waveforms from an identified CRF+ neuron. Bottom left,
excitation of the other population (Fig. 4m–o). Therefore, the cellular example CRF+ unit identified with ChR2. Top right, jitter of light-evoked
substrates of conditioned flight and freezing behaviour are poised to responses. Bottom right, waveform similarity of action potentials in and
inhibit each other through reciprocal inhibitory synaptic connections, out of light. c–h, Normalized activity (z score, bottom) and example raster
a circuit motif which would allow for rapid switching between behav- plots (top) of CEl CRF+ or SOM+ neurons. c, CRF+ neuronal responses
(n =​ 8, 7 mice, from 60 total units) to white noise during pre-exposure
ioural strategies20.
(day 1). d, CRF+ neuronal responses to white noise (n =​ 8) are enhanced
Recent data indicate that the CEA may act as a hierarchy generator on day 3 of the flight paradigm. e, CRF+ excitation to white noise is
between innate and learned fear responses through integration of sen- diminished (n =​ 7) in a neutral context. f, SOM+ neuronal responses
sory information21. Our results identify a novel competitive mechanism (n =​ 10, 7 mice, from 46 total units) to white noise during pre-exposure
in which shifting the balance between two classes of inhibitory neurons (day 1). g, SOM+ neurons are inhibited during white noise on day 3
within the CEA alters learned fear states. Our data demonstrate that (n =​  10). h, SOM+ cells (n =​ 8) are excited by white noise in the neutral
this process is crucial for action selection under stimulus-driven high- context. i, Normalized neuronal activity of identified units during
fear states; other CEA circuit mechanisms may account for expres- behavioural events. CRF+ cells are excited during flight (left, 1.9 s average
sion of defensive behaviours in response to ambiguous threat18,21 duration of behaviour) but not during freezing (left centre, 14.2 s average
duration of behaviour). SOM+ cells are inhibited during flight (right
(see Supplementary Discussion).
centre, 2.0 s average duration of behaviour) and excited during freezing
Monosynaptic rabies tracing revealed putative sources of excita- (right, 7.6 s average duration of behaviour). Horizontal lines in b indicate
tory input to CRF+ and SOM+ cells (Extended Data Fig. 6), which the mean. All other values are means ±​  s.e.m.
could convey information about context and threat level17,22. Increased
CRF+ cellular activity in response to these inputs then locally inhibits CEA cell types to influence action selection. Importantly, expression
SOM+ cells to reduce conditioned freezing. Conversely, excitation of of flight or freezing not only requires diminished neuronal activ-
SOM+ cells acts locally to dampen CRF+ cellular activity, and would ity of SOM+ or CRF+ cells, respectively, but also needs adequate
generate freezing responses via direct or indirect projections to the stimulus or context-driven input to these CEA neuronal subclasses.
periaqueductal grey14,16,17,23. Notably, our data suggests that CRF+ In addition, integration of the signals necessary to orchestrate flight
and SOM+ cells receive differential inputs from presynaptic regions is also probably carried out in the hypothalamus1 and periaqueductal
such as the basolateral amygdala and ventral hippocampus (Extended grey1,23,27,28, regions targeted by CRF+ neurons (Extended Data
Data Fig. 6a–e). Recent studies have demonstrated heterogeneity in Fig. 6f–k). Ultimately, integration of information related to threat is
neuronal coding and behavioural function amongst projections from probably carried out through both local and long-range interactions
these regions24–26, which suggests that they may impinge on specific at multiple sites.

0 0 M O N T H 2 0 1 7 | VO L 0 0 0 | NAT U R E | 3
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH LETTER

a AAV–flex–ChR2–eYFP b Conditioning (day 3) Figure 4 | CRF+ and SOM+ neurons form a reciprocal inhibitory circuit
or AAV–flex–GFP Fibre
CRF–ChR2 or SOM–ChR2
controlling defensive behaviour. a, Injection and implantation scheme
connector
4 × SCS
10 s 10 s for gain-of-function experiments. Mouse brain figure reproduced with
No light permission from ref. 31. b, Internally controlled behavioural protocol
design. c, Left, speed traces of the CRF–ChR2 group (n =​ 7) during light
CRF–ChR2 Light on
4 × SCS
and no-light trials. Right, speed traces from the SOM–ChR2 group (n =​  8)
Tone + light during light and no-light trials. d, Top, change in flight scores on day 3
CEA
BLA
Mouse lines: SOM–ChR2 Light on (CRF–ChR2, n =​ 7; CRF control, n =​  10; unpaired t-test; SOM–ChR2, n =​  8;
CRF-ires-cre 4 × SCS SOM–GFP control n =​  7; unpaired t-test). Bottom, change in freezing on day
SOM-ires-cre White noise
+ light
3 (CRF–ChR2 versus CRF–GFP control, unpaired t-test test; SOM–ChR2
versus SOM–GFP control, unpaired t-test). e, Whole-cell recordings from
c Tone White noise d Tone White

Change in Change in flight

onset onset 150 *** noise CEl target cells were performed while stimulating ChR2-expressing CRF+

score(%)
3 6 100
Tone White noise 50 ***
cells with blue light. f, Left, ChR2-expressing fibres from CRF+ cells. Right,
Speed (cm s–1)

Tone + light White noise 0 biocytin-filled SOM+ neuron identified with immunohistochemistry.
2 4 + light –50 Bottom, IPSCs from an example SOM+ neuron, evoked by photostimulation
40 * ** of the CRF+ network (individual traces in grey, corresponding average

freezing (%)
1 2
0 trace in black). g, Fraction of neurons receiving inhibitory input upon
0 0
–40
CRF
photostimulation of CRF+ cells (χ2 test, P <​ 0.01; Fisher’s exact test).
–10 –5 0 5 10 –10 –5 0 5 10 –80 SOM h, Average IPSC amplitudes from all recorded neurons (SOM+, n =​  23;
Timee (s) Time (s) Control
PKCδ​+, n =​  8; unidentified, n =​ 13; Kruskal–Wallis test, not significant).
e f g h i, Whole-cell recordings from CEl target neurons were performed while
ChR2 Biocytin SOM
SOM PKCδ stimulating ChR2-expressing SOM+ cells with blue light. j, Left, ChR2-
distribution (%))

100 2,000 Unident.

expressing fibres from SOM+ cells. Right, biocytin-filled CRF+ neuron
IPSC amplitude (pA)

CRF
CRF
ChR
hR2
hR
R2
2
ChR2
** **
SOM
or
20 μm 75 600 identified by tdTomato expression. Bottom, IPSCs from one CRF+ neuron,
PKCδ Light 50 400 evoked by photostimulation of the SOM+ network (individual traces in grey,
on
corresponding average trace in black). k, Fraction of CEA neurons receiving
Relative d

25 200
inhibitory inputs upon photostimulation of SOM+ cells, demonstrating
23/32

Whole-cell
13/17
200 pA

8/28

patch-clamp
patch-clamp
t h l
recording
rec
cording and
an 10 ms 0 0 preferential connectivity of SOM+ onto CRF+ neurons (χ2 test, P <​  0.01;
2

light stimulation
Fisher’s exact test). l, Average IPSC amplitudes from all recorded cells
i j ChR2
2 Biocytin k l CRF (CRF+, n =​  24; PKCδ​+, n =​  7; unidentified, n =​ 8; Kruskal–Wallis test, not
CRF PKCδ significant). m, Example traces from a SOM+ cell, demonstrating spike
(%)
%)

100 800
on (%

Unident.
mplitude (pA)

SOM
S OM *** suppression with ChR2-mediated excitation of the CRF+ network.
distribution
istributio

ChR2
ChR
Ch
hR2
hRR2
2 75 300
20 μm
n, Example traces from a CRF+ cell, demonstrating spike suppression with
IPSC amplitude

CRF
or
PKCδ
Lig
gh
Light
on
on 50 200 ChR2-mediated excitation of the SOM+ network. o, Left, spike probability
am

of SOM+ neurons is significantly reduced with concomitant CRF+ network

Relative d

25 10
100
24/31

excitation (two-tailed paired t-test). Right, CRF+ neuronal activity is

50 pA

7/28
8/17

10
0 mss 0 0 significantly diminished by concomitant SOM+ network activation
(Wilcoxon matched-pairs signed-rank test). Horizontal lines indicate means.
m CRF to SOM n SOM to CRF o
*** *** Box and whisker plots of in vitro experiments indicate median, interquartile
probability (%)

100
range, and 10th to 90th percentiles of the distribution. Crosses indicate
Spike

50 means. Box and whisker plots of behavioural experiments indicate median,

interquartile range, and 5th to 95th percentiles of the distribution. Crosses
20 mV

0
indicate means. All other values are means ±​  s.e.m. *​P <​  0.05; *​*​P <​  0.01;
C ent

C ent

ht t
h t

lig n
lig en

1s
*​*​*​P <​  0.001.
+ urre
C t
r

r
+ urr
ur

ur

Current Current Light

C

+ light

Survival in a volatile world is dependent on rapid selection of appro- 5. Tovote, P., Fadok, J. P. & Lüthi, A. Neuronal circuits for fear and anxiety. Nat. Rev.
priate action programs. Therefore, how discrete neuronal circuit Neurosci. 16, 317–331 (2015).
6. Davis, M. in The Amygdala (ed. J. P. Aggleton) Ch. 6, 213–288 (Oxford
elements interface to generate switches between behavioural states is of University Press, 2000).
prime importance. Previous experimental and theoretical work on the 7. Tsetsenis, T., Ma, X. H., Lo Iacono, L., Beck, S. G. & Gross, C. Suppression of
mechanisms of action selection in striatal circuits suggests an important conditioning to ambiguous cues by pharmacogenetic inhibition of the dentate
gyrus. Nat. Neurosci. 10, 896–902 (2007).
role for local recurrent inhibitory interactions20. Our study indicates 8. Perusini, J. N. & Fanselow, M. S. Neurobehavioral perspectives on the
that this circuit motif may represent a particularly effective and widely distinction between fear and anxiety. Learn. Mem. 22, 417–425 (2015).
used mechanism allowing for the rapid and flexible selection of beha­ 9. Maren, S. Neurobiology of Pavlovian fear conditioning. Annu. Rev. Neurosci. 24,
897–931 (2001).
vioural programs necessary for survival, and suggests that alterations in 10. LeDoux, J. E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184
the set points of such circuit switches could contribute to maladaptive (2000).
behaviour in anxiety disorders29,30. 11. Haubensak, W. et al. Genetic dissection of an amygdala microcircuit that gates
conditioned fear. Nature 468, 270–276 (2010).
Online Content Methods, along with any additional Extended Data display items and 12. Roberts, G. W., Woodhams, P. L., Polak, J. M. & Crow, T. J. Distribution of
Source Data, are available in the online version of the paper; references unique to neuropeptides in the limbic system of the rat: the amygdaloid complex.
these sections appear only in the online paper. Neuroscience 7, 99–131 (1982).
13. Swanson, L. W., Sawchenko, P. E., Rivier, J. & Vale, W. W. Organization of ovine
Received 30 March; accepted 5 December 2016. corticotropin-releasing factor immunoreactive cells and fibers in the rat brain:
an immunohistochemical study. Neuroendocrinology 36, 165–186 (1983).
Published online 25 January 2017.
14. Li, H. et al. Experience-dependent modification of a central amygdala fear
circuit. Nat. Neurosci. 16, 332–339 (2013).
1. Gross, C. T. & Canteras, N. S. The many paths to fear. Nat. Rev. Neurosci. 13, 15. Viviani, D. et al. Oxytocin selectively gates fear responses through distinct
651–658 (2012). outputs from the central amygdala. Science 333, 104–107 (2011).
2. Blanchard, D. C. & Blanchard, R. J. in Handbook of Anxiety and Fear Vol. 17 16. Ciocchi, S. et al. Encoding of conditioned fear in central amygdala inhibitory
Handbook of Behavioral Neuroscience (eds R. J. Blanchard, D. C. Blanchard, G. circuits. Nature 468, 277–282 (2010).
Griebel & D. Nutt) Ch. 2.4, 63–79 (Academic Press, 2008). 17. Penzo, M. A., Robert, V. & Li, B. Fear conditioning potentiates synaptic
3. Anderson, D. J. & Adolphs, R. A framework for studying emotions across transmission onto long-range projection neurons in the lateral subdivision of
species. Cell 157, 187–200 (2014). central amygdala. J. Neurosci. 34, 2432–2437 (2014).
4. Fanselow, M. S. Conditioned and unconditional components of post-shock 18. Gozzi, A. et al. A neural switch for active and passive fear. Neuron 67, 656–666
freezing. Pavlov. J. Biol. Sci. 15, 177–182 (1980). (2010).

4 | NAT U R E | VO L 0 0 0 | 0 0 M O N T H 2 0 1 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

19. Yu, K., Garcia da Silva, P., Albeanu, D. F. & Li, B. Central amygdala somatostatin
neurons gate passive and active defensive behaviors. J. Neurosci. 36,
6488–6496 (2016).
20. Nelson, A. B. & Kreitzer, A. C. Reassessing models of basal ganglia function and
dysfunction. Annu. Rev. Neurosci. 37, 117–135 (2014).
21. Isosaka, T. et al. Htr2a-expressing cells in the central amygdala control
the hierarchy between innate and learned fear. Cell 163, 1153–1164
(2015).
22. Penzo, M. A. et al. The paraventricular thalamus controls a central amygdala
fear circuit. Nature 519, 455–459 (2015).
23. Tovote, P. et al. Midbrain circuits for defensive behaviour. Nature 534, 206–212
(2016).
24. Beyeler, A. et al. Divergent routing of positive and negative information
from the amygdala during memory retrieval. Neuron 90, 348–361
(2016).
25. Xu, C. et al. Distinct hippocampal pathways mediate dissociable roles of
context in memory retrieval. Cell 167, 961–972.e16 (2016).
26. Ciocchi, S., Passecker, J., Malagon-Vina, H., Mikus, N. & Klausberger, T. Brain
computation. Selective information routing by ventral hippocampal CA1
projection neurons. Science 348, 560–563 (2015).
27. Bittencourt, A. S., Carobrez, A. P., Zamprogno, L. P., Tufik, S. & Schenberg, L. C.
Organization of single components of defensive behaviors within distinct
columns of periaqueductal gray matter of the rat: role of N-methyl-d-aspartic
acid glutamate receptors. Neuroscience 125, 71–89 (2004).
28. De Oca, B. M., DeCola, J. P., Maren, S. & Fanselow, M. S. Distinct regions of the
periaqueductal gray are involved in the acquisition and expression of defensive
responses. J. Neurosci. 18, 3426–3432 (1998).
29. Canteras, N. S. & Graeff, F. G. Executive and modulatory neural circuits of
defensive reactions: implications for panic disorder. Neurosci. Biobehav. Rev.
46, 352–364 (2014).
30. Pitman, R. K. et al. Biological studies of post-traumatic stress disorder. Nat. Rev.
Neurosci. 13, 769–787 (2012).
31. Paxinos, G. & Franklin, K. B. J. The Mouse Brain in Stereotaxic Coordinates.
(Academic, 2001).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTER RESEARCH
Supplementary Information is available in the online version of the paper.
Acknowledgements We thank N. Whittle for comments on the manuscript.
C. Ramakrishnan and K. Deisseroth provided intersectional viral vectors. M. S.
Esposito and S. Arber provided flex-synaptophysin-GFP viral vectors. Z. Josh
Huang provided the SOM-Flp mouse line. We thank J. Eglinger for providing
overlap analysis scripts and S. Bourke for imaging advice. J.P.F. was funded
by an EMBO LTF (952-2011) and a NARSAD Young Investigator Fellowship.
C.X. was funded by EMBO ALTF (1579-2010). S.K., L.M. and P.T. were funded
by a NARSAD Young Investigator Fellowship. All authors were supported by
the National Center of Competences in Research: ‘SYNAPSY — The Synaptic
Bases of Mental Diseases’ (financed by the Swiss National Science Foundation),
a SNSF core grant (to A.L.), an ERC Advanced Grant (to A.L.), and the Novartis
Research Foundation.
Author Contributions J.P.F. conceived, designed, performed and analysed
most of the experiments and wrote the manuscript. S.K. conceived, designed,
performed and analysed all in vitro experiments. M.M. contributed to project
conceptualization, and performed and analysed rabies tracings. J.C. analysed
single-unit data in relation to behaviour. C.X. generated rabies tracing tools,
and performed and analysed rabies tracing experiments. L.M. performed the
overlap analysis. P.B. performed experiments. K.B., C.M. and A.K. performed
experiments and analysed histology. P.T. and A.L. conceptualized the project
and wrote the manuscript. All authors contributed to the experimental design
and commented on the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial
interests. Readers are welcome to comment on the online version of the
paper. Correspondence and requests for materials should be addressed to
A.L. (andreas.luthi@fmi.ch) or P.T. (philip.tovote@fmi.ch).
Reviewer Information Nature thanks K. Tye and the other anonymous
reviewer(s) for their contribution to the peer review of this work.
0 0 M O N T H 2 0 1 7 | VO L 0 0 0 | NAT U R E | 5
 RESEARCH LETTER
METHODS
Animals. Male C57/Bl6J (Harlan/Envigo), Crf-ires-cre32, Som-ires-cre32, and
Pkcd-ires-cre11 mice aged 2–5 months were used for recording and behaviour. All
Cre driver lines were fully backcrossed to C57/Bl6J. For some in vitro experiments,
Crf-ires-cre mice were crossed to Som-ires-FlpO mice (Z. Josh Huang, Cold Spring
Harbour Laboratory) to generate CRF–Cre::SOM–Flp mice. For rabies tracing,
Crf-ires-cre and Som-ires-cre mice were crossed to a LSL-R26Tva-lacZ33 line to
gene­rate CRF–Cre::LSL–R26Tva–lacZ (CRF–Cre–TVA) and SOM–Cre::LSL–
R26Tva–lacZ (SOM–Cre–TVA) mice. All mice were individually housed on a 12 h
light/dark cycle during all experiments with ad libitum food and water. Behavioural
experiments were performed during the light cycle. All animal procedures were
performed in accordance with institutional guidelines and were approved by the
Veterinary Department of the Canton of Basel-Stadt.
Conditioned flight paradigm. Two different contexts were used for all behavioural
experiments. In context A, the mice were placed into a clear, cylindrical chamber
with a smooth floor. The chamber was cleaned with 1% acetic acid. Context
B contained a clear, square enclosure with an electrical grid floor (Coulbourn
Instruments) used to deliver footshock. This chamber was cleaned with 70%
ethanol. Both chambers contained an overhead speaker (model H12-01R,
Coulbourn Instruments) through which auditory stimuli were delivered at 75 dB.
Context B also contained a stimulus isolator (ISO-Flex, A.M.P.I.) for the delivery
of direct current (DC) shock. Auditory stimuli were generated using a System
3 RP2.1 real time processor, an SA1 stereo power amp, and RPvdsEx software
(all Tucker-Davis Technologies). Behavioural protocols were generated with
Radiant software (Plexon) to control all stimuli (CS, laser, shock and so on) via
TTL pulses with high temporal precision.
On day 1, following a 3 min baseline period, mice were presented with four
presentations of an auditory serial compound stimulus (SCS) of 20 s duration
in context A, with a 60 s average pseudorandom ITI (range 50–90 s). The SCS
was comprised of a 10 s pure tone period (500 ms, 7.5 kHz pips at 1 Hz), followed
immediately by a 10 s white noise period (500-ms pips at 1 Hz). The white noise
was random and composed of frequencies from 1–20,000 Hz. On day 2 and day 3,
after a 3 min baseline period, mice were conditioned in context B by five pairings
of the auditory stimulus with a 1 s, 0.9 mA DC footshock, with a 180 s average
pseudorandom ITI (range 150–210 s). The shock was applied immediately at the
offset of the last pip. For retrieval tests, mice were placed back into context A and,
following a 3 min baseline period, were presented with the auditory stimulus four
times with an average ITI of 90 s (range 60–120 s). Extinction of conditioned flight
behaviour was performed in context B. Following a 3 min baseline period, mice
were presented with 16 presentations of the SCS with an average pseudorandom
ITI of 90 s (range 60–100 s). For the no shock control group, this paradigm was
unaltered except for the removal of the electric footshock on day 2 and day 3. In
initial experiments that led to the development of the SCS, the auditory stimuli
were either a pure 7.5 kHz tone or white noise.
Quantification of behaviour. All behavioural sessions were recorded to video
using Cineplex software (Plexon) and mice were tracked using contour tracking.
Freezing behaviour was scored from the video recording by an observer blind to
the condition, using the assistance of a frame-by-frame analysis of pixel changes
(Cineplex Editor, Plexon). By calibrating the known size of the chambers with
the pixel dimensions of the camera to determine a calibration coefficient, speed
(cm s−1) was extracted using the animal’s centre of gravity. Jump escape behaviours
were scored manually from video files. The flight score was calculated by dividing
the average speed during each CS by the 10 s of pre-CS speed (speedCS/speedBL)
and then adding 1 point for each escape jump. A flight score of 1 therefore indicates
no change from the pre-CS period.
Surgery. Mice were deeply anaesthetized with 5% isoflurane (Attane, Provet) in
oxygen-enriched air (Oxymat 3, Weinmann), given an intraperitoneal injection
of 1 mg kg−1 meloxicam (Metacam, Boehringer Ingelheim), and then fixed into
a stereotaxic alignment instrument (Model 1900, Kopf Instruments). During
surgery, mice were maintained on 1–2.5% isoflurane. Prior to scalp incision, a local
injection of 0.1 ml ropivacain (Naropin, AstraZeneca) was made subcutaneously.
Mice were maintained at a core temperature of 36 °C using a feedback-controlled
DC temperature controller (FHC).
For selective expression of inhibitory or excitatory opsins, Cre-dependent
recombinant adeno-associated viral vectors (AAV) were injected into the central
amygdala (CEA). AAVs (approximately 0.25 μ​l per hemisphere) were delivered
into the CEA using pulled glass pipettes (tip diameter 10–20 μ​m) connected to a
Picospritzer III microinjection system (Parker Hannifin Corporation). CEA coor-
dinates were: 1.2–1.3 mm posterior to bregma, ±​2.9–3.0 mm lateral to the midline
and 3.9–4.1 mm below the cortical surface.
For optogenetic manipulations of behaviour, animals were bilaterally implanted
with custom-built optic fibre connectors (fibre: 0.48 numerical aperture, 200 μ​m
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
diameter, Thorlabs) two weeks after vector injection using the aforementioned
methods. Optic fibre tips were lowered to 100–300 μ​m above the CEA. Implants
were fixed to the skull with miniature screws (P.A. Precision Screws), cyanoacrylate
glue (Ultra Gel, Henkel) and dental cement (Paladur, Heraeus). Mice had two
additional weeks of recovery to allow for adequate opsin expression.
For in vivo single-unit recordings, one hemisphere was implanted with an
optrode, consisting of an optic fibre with an attached 16-wire electrode, two
weeks after vector injection. The electrodes tips were cut at an angle to protrude
approximately 300–500 μ​m beyond the fibre end. Electrodes were made of indi-
vidually insulated, gold-plated nichrome wires (13 μ​m inner diameter, impedance
70–100 kΩ​, Sandvik). The end of each wire was de-insulated and attached to an
18-pin connector (Omnetics). The implant was fixed to the skull as described
above.
Viruses. For Cre-dependent expression of archaerhodopsin (Arch), we used
rAAV2/5 CBA::flex–Arch–GFP. For Cre-dependent expression of channelrho-
dopsin-2 (ChR2), we used rAAV2/5 EF1a::flex–ChR2(H134R)–eYFP. Control
mice for optogenetic experiments were mixed Cre+ mice from the Crf-ires-cre
and Som-ires-cre colonies that were injected with rAAV2/1 CAG::flex–GFP. To label
CRF+ somata for in vitro experiments, CRF–Cre::SOM–Flp mice were injected
with rAAV2/1 CAG::flex–TdTomato. All of the above mentioned viruses were
obtained from the University of Pennsylvania Vector Core. For Flp-dependent
expression, we injected rAAVdj hSyn::CreOff/FlpOn–ChR2–eYFP34, or rAAV5
hSyn::CreOff/FlpOn–mCherry (Deisseroth laboratory, Stanford, produced by
Vector Biolabs). For rabies tracing, custom-made rAAV2/7–EF1α​–flex–rabiesG
(UPenn Vector Core) and rabiesΔ​G–GFP–EnvA viruses (C. Xu, FMI) were used.
For axonal tracing experiments, rAAV2/9 CAG::flex–synaptophysin–GFP was
used (S. Arber, FMI, produced by Vector Biolabs).
Optogenetic manipulations of behaviour. For optogenetic manipulations during
behaviour, the implanted fibres were connected to a custom-built laser bench (Life
Imaging Services) using an acousto-optic modulator (AA Opto-Electronic) to
control laser intensity (lasers: MBL473, 473 nm wavelength and MGL593.5,
593.5 nm wavelength, CNI Lasers). The laser was connected from the port to the
implanted fibre via a custom-made patch cable (Thorlabs). Laser power at the
fibre tip was measured before every subject with an optical power and energy
meter (PM100D, ThorLabs). Connectors were tested for coupling efficiency before
implantations and laser power at the fibre tip for behavioural manipulations was
adjusted to reach an approximate irradiance value of 11 mW mm−2.
Cre+ mice of the same strain but from different litters were used in individual
experiments. Mice were allocated to experimental groups without pre-determined
criteria and experimental subjects had unique identifiers that were later used to
indicate genotype and group assignment. Optogenetic manipulations of condi-
tioned behaviour were always performed with an internally controlled design in
which mice underwent eight trials, with alternating light-off and light-on con-
ditions. To inhibit CEA neuronal subpopulations during conditioned flight, the
yellow light was switched on 500 ms before the onset of the first white noise pip
and remained on until the end of the last white noise pip (10 s in total). To inhibit
SOM+ cells during the tone on day 3, the yellow light was switched on 500 ms
before the onset of the first tone pip and was switched off immediately following
the last tone pip (10 s total duration). For behavioural measurements in naive mice
of the Arch groups, the yellow light was switched on for 10 s, four times, with an
ISI of 60 s. For ChR2 excitation of CRF+ cells, we used 10 ms pulses of blue light
at 10 Hz. On day 3, the light was switched on 500 ms before the first pure tone pip
and the pulse train continued for 10 s. On day 4, the light was switched on at the
onset of the first pure tone pip and remained on for the entire SCS (20.5 s total).
For ChR2 excitation of SOM+ cells, we used 5-ms pulses at 50 Hz (ref. 14). To test
for effects in naive mice, the blue light pulse train (appropriate for each group) was
on for 10 s, four times, with an ISI of 60 s.
Single-unit recordings. Single-unit spike sorting was performed using the Off-
Line Spike Sorter software (Plexon) as previously described16,35,36. In brief, princi-
pal components were calculated for unsorted waveforms and plotted in principal
component space. Clusters were manually defined, template waveforms were saved,
and template-matching was used to verify unit identity across sessions. Sort quality
was quantified using the J3 statistic and Davies–Bouldin validity index, as previ-
ously described16,35,36. Additionally, we determined unit stability by quantifying the
similarity of waveform shapes using linear correlation (r) values16,35,36.
Neuronal activity is reported either as frequency or normalized activity (z-score
transformation). z scores used to determine CS-induced neuronal activity changes
were calculated for each neuron by subtracting the population mean (the average
baseline firing rate in the 300 ms pre-stimulus period across all trials and pips) from
each individual raw value (50-ms bins across all trials and pips), and then dividing
this difference by the standard deviation of the population mean. z scores for each
neuron were then averaged across cells for each population. Any statements of

significance were made only if the average activity of an identified population
reached a ±​1.645 change in z score for at least one bin during the period of interest.
To evaluate unit activity relative to behaviour, we normalized to the episodes of
freezing and flight behaviours. Because the duration of those behaviours is variable,
the time periods before, during, and after freezing and flight behaviours were each
divided into equal numbers of bins and spikes were sorted into these bins. The
resulting activity was then normalized to the pre-freezing or pre-flight period
using a z-score transformation.
Optical identification of units. For optogenetic identification of CRF+ or SOM+
neurons, we used pulses of either blue or yellow light (to activate ChR2 or Arch,
respectively). We used 300-ms pulses, 60–120 times, with a 2 s inter-pulse interval,
at 10–20 mW light power at the fibre tip. Units were considered as light responsive
if they showed significant, time-locked (<​10 ms) changes in neuronal activity upon
illumination36. For units identified with ChR2, we calculated the jitter of first light-
evoked spikes by calculating the difference between samples and then dividing by
the number of samples (minus 1). To determine the onset of inhibition in neu-
rons with low baseline firing rates, we used change-point analysis (Change Point
Analyzer 2.0, Taylor Enterprises Inc.). As described previously36, this identifies
the time point exhibiting a significant change in neuronal activity relative to the
preceding time points. For all identified units, we calculated linear correlation (r)
values for spontaneous and light-evoked spikes to quantitatively determine the
similarity of their waveform shapes. Note that for two SOM+ cells identified using
Arch, there were not enough spikes inside the light period to calculate the r value.
In vitro electrophysiology. Four weeks after virus delivery, coronal brain slices
were prepared from CRF-ires-cre or CRF– Cre::SOM –Flp mice injected with rAAV
2/5 EF1a::flex–ChR2(H134R) –eYFP or rAAVdj hSyn::CreOff/FlpOn–ChR2–eYFP
and rAAV2/1 CAG::flex–tdTomato, respectively. For some experiments, CRF–
Cre::SOM–Flp mice injected with rAAV 2/5 EF1a::flex–ChR2(H134R) –eYFP
and rAAV5 hSyn::CreOff/FlpOn–mCherry were used. Mice were deeply anaes-
thetized by intraperitoneal injection of ketamine/medetomidine (250 mg kg−1
and 2.5 mg kg−1 bodyweight, respectively) and transcardially perfused with ice-
cold ACSF (in mM: 124 NaCl, 2.7 KCl, 26 NaHCO3, 1.25 NaH2PO4, 2.5 glucose,
50 sucrose, 0.1 CaCl2, 6 MgCl2, 3 kynurenic acid, oxygenated with 95% O2/5%
CO2, pH 7.4). The brain was quickly removed from the skull and coronal brain
slices (300 μ​m) containing the CEA were cut in ice-cold perfusion ACSF using
a vibratome (HM 650 V, Microm) equipped with a sapphire blade. Brain slices
were allowed to recover in an interface chamber with recording ACSF (in mM:
126 NaCl, 2.7 KCl, 26 NaHCO3, 1.25 NaH2PO4, 18.6 glucose, 2.25 ascorbic acid,
2 CaCl2, 1.3 MgCl2, oxygenated with 95% O2/5% CO2, pH 7.4) for 45 min at 37 °C
and subsequently kept at room temperature (20–22 °C) until start of recordings.
Sections were transferred to a submerged recording chamber on an upright
microscope (BX50WI, Olympus) and constantly superfused with recording ACSF
(2–4 ml min−1; as above, except: 2.5 mM CaCl2, 10 μ​M CNQX (6-cyano-7-nitro-
quinoxaline-2,3-dione), 10 μ​M CPP (3-((R)-2-carboxypiperazin-4-yl)-propyl-1-
phosphonic acid)) at 35 ±​ 1 °C. eYFP- and tdTomato/mCherry-expressing neurons
were visualized using epifluorescence and a 40×​objective (LumPlanFl 40×​/0.8,
Olympus).
Voltage clamp recordings from CEA neurons were acquired in whole-cell mode
at a holding potential of −​70 mV using 3–5 MΩ​glass pipettes filled with inter-
nal solution (in mM: 110 CsCl, 30 K-gluconate, 1.1 EGTA, 10 HEPES, 0.1 CaCl2,
4 Mg-ATP, 0.3 Na-GTP, 4 QX-314 chloride and 0.4% biocytin, pH 7.3). ChR2-
expressing neurons were photostimulated using a blue LED (465 nm, Plexon, LED-
driver LD-1) connected to an optical fibre, which was positioned above the CEA.
Blue light pulses of 10 mW with 10 ms duration were applied at a frequency of 1 Hz.
In some slices, 100 μ​M picrotoxin was administered with the recording ACSF for
the last cell recorded from. Spike suppression was tested in current clamp mode
(internal solution in mM: 130 K-methylsulfate, 10 HEPES, 10 Na-phosphocreatine,
4 Mg-ATP, 0.3 Na-GTP, 5 KCl, 0.6 EGTA and 0.4% biocytin, pH 7.3). Spikes were
evoked from a holding potential of −​60 mV with current injections of 20–60 pA
for 500 ms, and the same current step was subsequently paired with blue light
stimulation to activate ChR2-expressing neurons. Spike probability of individual
neurons with and without blue light stimulation was calculated from ten repeated
trials, except for one SOM+ cell in which there were nine repeated trials. Data were
acquired with a Multiclamp 700A amplifier, Digidata 1440A A/D converter and
pClamp 10 software (all Axon instruments) at 20 kHz and filtered at 4 kHz (voltage
clamp) or 10 kHz (current clamp). Series resistance was monitored continuously
and experiments were discarded when exceeding 30 MΩ​. In each slice, 2–6 CEA
neurons surrounded by ChR2–eYFP-positive fibres were recorded and filled with
biocytin for post hoc immunohistochemical identification of cell type after elec-
trophysiology. Outside-out patches were pulled at the end of each recording and
slices were fixed for 1 h in 4% paraformaldehyde in PBS for subsequent immu-
nohistochemistry. All chemicals were purchased from Sigma-Aldrich except for
CNQX, CPP and QX-314 chloride (from Tocris Bioscience).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTER RESEARCH
Immunohistochemistry related to in vitro physiology. After fixation for 1 h in
4% paraformaldehyde in PBS, 300-μ​m brain slices were stored in PBS overnight
(4 °C). The next day, free-floating sections were washed with PBS three times before
incubation in blocking solution (3% normal goat serum (vol/vol), 1% bovine serum
albumin (BSA, vol/vol) and 0.5% Triton X-100 (vol/vol) in PBS) for 2–3 h at room
temperature (20–22 °C). Afterwards, slices were incubated for 48 h at 4 °C in carrier
solution (same as above) with a combination of the following antibodies: rat anti-
SOM (1:500; Millipore, MAB354), mouse anti-PKCδ​(1:500; BD Biosciences,
610398), rabbit anti-GFP (1:1,000; Life Technologies, A11122), chicken anti-RFP
(1:500; Millipore, AB3528), mouse anti-bassoon (1:500, Enzo Life Sciences, ADI-
VAM-PS003). Sections were washed with 0.1% Triton X-100 in PBS three times
before adding secondary antibodies (goat anti-rat Alexa Fluor 647, 1:500, A21247;
goat anti-mouse Alexa Fluor 647 or 568, 1:750, A31625 or A11031; goat anti-rabbit
Alexa Fluor 488, 1:750, A27034; goat anti-chicken Alexa Fluor 568, 1:500, A11041;
all Life Technologies) and streptavidin conjugated to Alexa Fluor 405 (1:1,000;
Life Technologies, S32351) in carrier solution for 24 h at 4 °C. Finally, slices were
washed in PBS four times, mounted and coverslipped, and imaged using a confocal
microscope (LSM710, Carl Zeiss AG) with 63×​objective and twofold digital zoom.
Only reconstructed cells were included in the data analysis. Recordings from cells
immuno-negative for SOM/mCherry, PKCδ​or tdTomato were summarized in an
‘unidentified’ group.
Histology and immunohistochemistry for optogenetics and single-unit recordings.
Mice were deeply anaesthetized with tribromoethanol (Avertin; 0.3 g kg−1) and an
electrolytic lesion was made at the electrode tip by applying 0.1 μ​A for 10 s (with a
polarity switch) to one of the electrode wires. Mice were then perfused transcardially
with PBS followed by 4% paraformaldehyde in PBS. Brains were removed and
post-fixed overnight at 4 °C. 80-μ​m coronal brain slices were cut with a vibratome
(VT1000 S, Leica) and stored in PBS. Standard immunohistochemical procedures
were performed on free-floating brain sections. Sections were blocked in 5% normal
goat serum (in 0.3% Triton X-100 in PBS; NGS) for 2 h at room temperature
(20–22 °C), followed by overnight incubation with primary antibodies (rabbit anti-
GFP, 1:1,000; Life Technologies, A11122) in NGS at 4 °C. The next day, slices were
washed 3×​for 10 min in 0.3% Triton X-100, and then incubated with secondary
antibodies (goat anti-rabbit Alexa Fluor 488, 1:1,000, A27034; Life Technologies)
in NGS for 2 h at room temperature. Slices were washed for 10 min in PBS, given a
10 min exposure to 4′​,6-diamidin-2-phenylindol (DAPI, 1:10,000, Sigma-Aldrich),
and then washed three times for 15 min in PBS. Slices were mounted on gelatin-
coated glass slides, coverslipped and imaged using an Axioscan Z1 (Carl Zeiss
AG). To quantify the number of opsin-expressing somata in the CEl versus CEm,
three sections around the optic fibre tip in each hemisphere were scanned with
a 10×​air objective using a laser-scanning confocal microscope (LSM700, Carl
Zeiss AG). z stacks of multi-tiled regions of interest were acquired. Manual cell
counting of fluorophore-expressing somata was then performed and values were
averaged.
Fibre tips and electrode lesions were located and matched against a mouse brain
atlas (Paxinos and Franklin, 2001). Mice were included into the behavioural analysis
if they had bilateral expression in the CEA and if the fibre tips were no more than
300 μ​m away from the CEA.
Overlap analysis. To detect overlap between CRF+ and PKCδ​+ or SOM+ neurons,
CRF-ires-cre mice were injected with rAAV2/1 CAG::flex–GFP to mark CRF+
somata. For quantification, serial vibratome sections (80 μ​m) were obtained from
two and three animals for PKCδ​and SOM, respectively. PKCδ​+ or SOM+ cells
were visualized by immunostaining using a primary antibody against PKCδ​ raised
in mouse (1:500, BD Transduction Laboratories, 610398) and SOM raised in rat
(1:500, Millipore, MAB354). They were detected by Alexa Fluor 647-conjugated
secondary antibodies (1:1,000, Invitrogen, goat anti-mouse Alexa Fluor 647,
catalogue number A-20990 and goat anti-rat Alexa Fluor 647, catalogue number
A-21248). For image acquisition, sections (PKCδ​, n =​  4; SOM, n =​ 6) were scanned
with a 20×​air objective (Plan-Apochromat 20×​, NA  =​ 0.8) using a laser-scan-
ning confocal microscope (LSM700, Carl Zeiss AG). z stacks of multi-tiled regions
of interest were then acquired. Individual tiles were stitched using the stitching
plugin37 in Fiji (ImageJ software) and downsampled (factor: 0.5). After down-
sampling, we applied a segmentation classifier using the 3D Weka segmentation
classifier trained in Fiji and probability maps were obtained. The probability maps
were then converted into cell segmentation masks using a Python script in Fiji.
The intensity of each individual cell was measured using the 3D manager plugin38.
Monosynaptic rabies tracing of inputs to CRF + and SOM + neurons.
A mixture of rabies∆​G –mCherry–EnvA, or rabies∆​ G –GFP–EnvA, and
AAV2/7–flex–rabiesG (3:2; 500 nl total volume) was injected into the CEA
of CRF–Cre–TVA and SOM–Cre–TVA mice. In other animals, we first
injected AAV2/7–flex–rabiesG, followed by an injection of rabies∆​G–GFP–
EnvA a week later. As the two strategies yielded comparable results, the data
were pooled together. One week after the rabies virus was injected, mice
 RESEARCH LETTER
were transcardially perfused with PBS, followed by 4% paraformaldehyde
in PBS. Brains were post-fixed overnight at 4 °C. Brains were cut into 80-μ​m
coronal slices with a vibratome (Leica). Free-floating sections were rinsed in PBS.
Sections were then incubated in blocking solution (3% bovine serum albumin and
0.5% Triton X-100 in PBS; BSA) containing a guinea pig anti-rabies glycoprotein
(rabG) antibody for 48 h at 4 °C (1:500, custom order, Eurogentec). Subsequently,
sections were washed with PBS three times for 5 min and then incubated for 4 h at
room temperature with goat anti-guinea pig Alexa Fluor 647 (1:500; Invitrogen,
A-21450) in BSA. Finally, immunolabelled sections were rinsed three times with
PBS, mounted on gelatin-coated slides, dehydrated and coverslipped. Individual
brain slices were imaged with a LSM700 confocal microscope (Carl Zeiss AG).
Quantitative analysis of rabies tracing results. Brain sections were scanned by an
AxioScan (Carl Zeiss AG) with a 5×​objective. Images for individual brain sections
were sorted in the correct order, aligned with the TrackEM plugin for Fiji (Fiji 1.48),
and then exported into Imaris software (Imaris 7.3.1) for analysis. The contours
for different brain structures were manually drawn in Imaris using a mouse brain
atlas for reference. Somata were semi-automatically detected as spots in Imaris
(automated detection followed by manual verification to eliminate false positive
and false negative results). Finally, the total number of cells for a given brain area
was exported from Imaris and quantified.
Anterograde tracing of axons from CRF+ neurons. To determine projection
targets of CRF+ neurons, CRF-ires-cre mice were injected into the CEA with an
AAV carrying a Cre-conditional synaptophysin conjugated to GFP to observe
axon terminals (S. Arber, FMI). Mice were perfused, and tissue was collected and
processed as described above (see Histology and immunohistochemistry for opto-
genetics and single-unit recordings). Sections were stained against GFP, with a
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
DAPI counterstain, as described above. Images were acquired with a laser-scanning
confocal microscope (LSM700, Carl Zeiss AG).
Statistics. All data were statistically analysed using Prism 6 (GraphPad Software,
Inc.). All data were checked for normal distribution using the Shapiro–Wilk
normality test (α =​ 0.05), and the appropriate parametric, or non-parametric,
test for significance was run. No statistical methods were used to predetermine
sample size.
Data availability. The data sets generated during and/or analysed during the
current study are provided as Source Data, or are available from the corresponding
author on reasonable request.
32. Taniguchi, H. et al. A resource of Cre driver lines for genetic targeting of
GABAergic neurons in cerebral cortex. Neuron 71, 995–1013 (2011).
33. Seidler, B. et al. A Cre-loxP-based mouse model for conditional somatic gene
expression and knockdown in vivo by using avian retroviral vectors. Proc. Natl
Acad. Sci. USA 105, 10137–10142 (2008).
34. Fenno, L. E. et al. Targeting cells with single vectors using multiple-feature
Boolean logic. Nat. Methods 11, 763–772 (2014).
35. Herry, C. et al. Switching on and off fear by distinct neuronal circuits. Nature
454, 600–606 (2008).
36. Wolff, S. B. et al. Amygdala interneuron subtypes control fear learning through
disinhibition. Nature 509, 453–458 (2014).
37. Preibisch, S., Saalfeld, S. & Tomancak, P. Globally optimal stitching of
tiled 3D microscopic image acquisitions. Bioinformatics 25, 1463–1465
(2009).
38. Ollion, J., Cochennec, J., Loll, F., Escudé, C. & Boudier, T. TANGO: a generic tool
for high-throughput 3D image analysis for studying nuclear organization.
Bioinformatics 29, 1840–1841 (2013).
 LETTER RESEARCH

Extended Data Figure 1 | See next page for caption.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH LETTER
Extended Data Figure 1 | Data from all mice presented in Fig. 1, and
single CS conditioned flight and no-shock control groups. a, The data
from Fig. 1d (day 3, conditioning) represented as a line plot to illustrate
changes in each mouse. Left, the number of escape jumps (n =​  20,
Friedman test, P <​ 0.0001, Dunn’s multiple comparisons test). Middle,
flight scores for each mouse. The pre-SCS period is used to calculate
the flight score and is therefore plotted as a value of 1 (n =​  20, Wilcoxon
matched-pairs signed-rank test). Right, freezing values (n =​  20; one-way
repeated-measures ANOVA, F =​  56.82, P <​ 0.0001, Tukey’s multiple
comparisons test). b, The data from Fig. 1e (day 4, extinction) represented
as a line plot to illustrate changes in each mouse. Left, flight scores for
each mouse. The pre-SCS period is used to calculate the flight score
and is therefore plotted as a value of 1 (n =​ 12; Wilcoxon matched-pairs
signed-rank test). Right, freezing values, including the pre-SCS period,
during the first block of four trials of extinction (n =​ 12; Friedman test,
P <​ 0.01, Dunn’s multiple comparisons test). c, The data from Fig. 1f (day
4, retrieval) represented as a line plot to illustrate changes in each mouse.
Left, flight scores for each mouse. The pre-SCS period is used to calculate
the flight score and is therefore plotted as a value of 1 (n =​  8; two-tailed
paired t-test). Right, freezing values, including the pre-SCS period (n =​  8;
one-way repeated-measures ANOVA, F =​  49.55, P <​  0.0001, Tukey’s
multiple comparisons test). d, Schematic of the paradigm used to elicit
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
flight in response to a single CS. e, Two groups of mice were subjected to
the flight paradigm and either a pure tone (n =​ 10) or white noise (n =​  10)
was paired with a 1 s shock during conditioning. f, Flight scores for each
group, across days. While active behaviour developed in both groups, the
flight response was significantly greater in the white noise group on day
3 on the first two trials (two-way repeated-measures ANOVA, cue ×​ trial
interaction, F(17,153) =​  1.81, P <​ 0.05, Bonferroni’s multiple comparison
test). g, Freezing responses for each group, across days. Freezing in
the pure tone or white noise group was similar on all trials (two-way
repeated-measures ANOVA, cue ×​trial interaction, F(17,153) =​  1.62,
P =​  0.06). h, Schematic of behavioural paradigm used to test for possible
aversive nature of white noise. i, The SCS was presented but was never
paired with footshock. j, Left, flight was not observed in the conditioning
context in the absence of pairing with footshock (n =​  10, two-tailed
paired t-test). Right, freezing values remained at low baseline values in
the absence of pairing with footshock (n =​ 10, Friedman test). k, Left,
flight was not observed in the neutral context (n =​ 10, two-tailed paired
t-test). Right, freezing values were decreased, not increased, in response
to the SCS (n =​ 10, one-way ANOVA, F =​  9.45, P <​ 0.01, Tukey’s multiple
comparisons test). Horizontal lines denote the mean. Values in f and g are
means ±​  s.e.m. *​P <​  0.05; *​*​P <​  0.01.
 LETTER RESEARCH

Extended Data Figure 2 | See next page for caption.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH LETTER
Extended Data Figure 2 | Data related to Fig. 2, and loss-of-function for
PKCδ groups, effects of optical inhibition on naive mice, and overlap
analysis. a, Positions of fibre tips for loss-of-function experiments.
Placements were determined using standard histological techniques
and lesions were matched to a mouse brain atlas. Mouse brain figure
reproduced with permission from ref. 31. b, Injection and implantation
strategy for identified single-unit recordings used to demonstrate Arch-
mediated inhibition. c, Single-unit recordings from identified CEl units
in behaving mice. Units were isolated and identified as described (Fig. 3,
Extended Data Fig. 3 and Methods). The light was switched on four times
for 10 s. Shown for each group is an example raster above and the average
peri-event histogram of firing rate below. The light inhibited neuronal
activity in all three cell types. d, Expression of Arch in CEl PKCδ​+ neurons.
Bregma −​1.34 mm caudal. Scale bar, 250 μ​m. e, Average speed traces of
PKCδ​–Arch mice (n =​ 8) in the conditioned flight paradigm, illustrating
that there was no effect of the light on conditioned flight (see Fig. 2 for
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
protocol). f, Average speed traces from naive mice given four, 10-s light
stimulations (GFP control, n =​  8;CRF–Arch, n =​  7; SOM–Arch, n =​  11;
PKCδ​–Arch, n =​  11). g, Flight and freezing values for all mice shown in
Fig. 2, plus the PKCδ​–Arch group (flight scores: two-tailed paired t-test
for GFP control, SOM–Arch, and PKCδ​–Arch; Wilcoxon matched-pairs
signed-rank test for CRF–Arch; freezing scores: Wilcoxon matched-pairs
signed-rank test for GFP control and SOM–Arch; two-tailed paired t-test
for CRF–Arch and PKCδ​–Arch). Horizontal lines indicate the mean.
h, Flight and freezing values for naive mice given four, 10-s light
stimulations (Wilcoxon matched-pairs signed-rank test for freezing
values). Horizontal lines indicate the mean. i, Example images from
overlap analysis between CRF+ neurons and SOM+ or PKCδ​+ neurons in
the CEA. Arrowhead indicates a rare example of overlap between CRF+
and PKCδ​+. Scale bar, 10 μ​m. j, Percentage of overlap. (PKCδ​+, n =​  4
sections from two mice; SOM+, n =​ 6 sections from three mice). Values are
means ±​  s.e.m. *​P <​  0.05.
 LETTER RESEARCH

Extended Data Figure 3 | See next page for caption.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH LETTER
Extended Data Figure 3 | Data related to Fig. 3, and neuronal responses
of identified units to tone on day 3. a, Positions of electrode lesions
and fibre tips for data presented in Fig. 3. Electrolytic lesions were made
at the tip of one wire (see Methods) and placements were determined
using standard histological techniques. Lesions were matched to a
mouse brain atlas. Mouse brain figure reproduced with permission
from ref. 31. b, Raster plots and peri-event histograms of firing rate
for example neurons from each condition illustrating Arch-mediated
inhibition, or ChR2-mediated excitation, of CEl CRF+ and SOM+ cells
(an example from the CRF–ChR2 group is presented in Fig. 3). c, Latency
to activation/inhibition plotted against z scores for all CRF+ and SOM+
neurons. Average latency: CRF+ =​  5.25  ±​ 0.88 ms; SOM+ =​  4.6  ±​ 0.9 ms.
d, Left, waveforms of action potentials were unchanged by light for all
unidentified units recorded in Arch-expressing mice with identified units.
Centre, firing rates of unidentified units before, during, and after light,
illustrating that the light pulses had no effect on unidentified unit activity.
Right, z-score transformation of unidentified unit activity during and
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
after light pulses. e–j, Normalized activity (z score, bottom) and example
raster plots (top) of CEl CRF+ or SOM+ neurons to the tone component
of the SCS. e, Neuronal response to the tone component for CRF+ neurons
(n =​ 8) during pre-exposure (day 1). Pre-SCS firing rate: 4.5 ±​  1.3 spikes
per s. f, CRF+ neurons were not significantly activated by the pure tone on
day 3 of the flight paradigm (n =​ 8). Pre-SCS firing rate: 5.5 ±​ 2 spikes per s.
g, There was no significant activation of CRF+ cells to the tone in the
neutral context (n =​ 7). Pre-SCS firing rate: 5.4 ±​ 2.3 spikes per s. h, SOM+
neuronal activity (n =​ 10) was not significantly changed in response to the
tone during the pre-exposure session (day 1). Pre-SCS firing rate: 2.7 ±​  0.6
spikes per s. i, Neuronal responses of SOM+ neurons to the tone on day 3
of the flight paradigm (n =​ 10). Pre-SCS firing rate: 5.8 ±​ 1.3 spikes per s.
These higher firing rates are consistent with the pre-SCS contextual
freezing observed on this day. j, SOM+ cells (n =​ 8) are excited by the tone
in the neutral context. Pre-SCS firing rate: 2.1 ±​ 0.4 spikes per s. Values are
means ±​  s.e.m.
 LETTER RESEARCH

Extended Data Figure 4 | See next page for caption.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH LETTER
Extended Data Figure 4 | Data related to Fig. 4, and effects of optical
excitation on naive mice and CRF–ChR2 mice during retrieval.
a, Top, positions of fibre tips for CRF+ and SOM+ gain-of-function
groups. Placements were determined using standard histological
techniques and lesions were matched to a mouse brain atlas. Mouse brain
figure reproduced with permission from ref. 31. b, Single-unit recordings
from identified CEl units in behaving mice. Units were isolated and
identified as described (see Fig. 3, Extended Data Fig. 3 and Methods).
The light was switched on four times for 10 s (see Methods). Shown for
each group is a raster plot of an example neuron above and a peri-event
histogram of the average firing rate below. The light protocols were
effective in exciting both neuronal subclasses. c, Average speed traces
from naive mice given four, 10-s light stimulations (CRF control, n =​  10;
CRF–ChR2, n =​ 7; SOM control, n =​  7; SOM–ChR2, n =​ 7). Blue light
caused a decrease in speed in the SOM–ChR2 group. d, Flight and freezing
values for naive mice given four, 10-s light stimulations. ChR2-mediated
excitation of SOM+ cells induced freezing behaviour (two-tailed paired
t-test, CRF–ChR2 and SOM–ChR2; Wilcoxon matched-pairs signed-rank
test, CRF control and SOM control). Horizontal lines indicate the mean.
e, Flight and freezing values for individual CRF control (left) and
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
CRF–ChR2 (right) mice shown in Fig. 4d (two-tailed paired t-test for
CRF control freezing, CRF–ChR2 flight, CRF–ChR2 freezing. Wilcoxon
matched-pairs signed-rank test for CRF control flight). Horizontal lines
indicate the mean. f, Flight and freezing values for individual SOM control
(left) and SOM–ChR2 (right) mice shown in Fig. 4d (two-tailed paired
t-test). Horizontal lines indicate the mean. g, The CRF–ChR2 (n =​  7)
and GFP control (n =​ 10) groups were subjected to the conditioned flight
paradigm. They were placed into the neutral context on day 4 (retrieval)
and the SCS was played eight times. On four of the trials, the blue light
was turned on during the entire SCS. Left, flight and freezing values
for individual mice in the CRF control group (flight scores: Wilcoxon
matched pairs signed-rank test; freezing scores: two-tailed paired t-test).
Middle, flight and freezing values for individual mice in the CRF–ChR2
group (two-tailed paired t-test). Horizontal lines indicate the mean.
Right, average change in freezing and flight between light and no-light
conditions. CRF+ excitation significantly reduced conditioned freezing
(two-tailed unpaired t-test). Box and whisker plots indicate median,
interquartile range, and 5th to 95th percentiles of the distribution. Crosses
indicate means. All other values are means ±​ s.e.m., except as indicated.

Extended Data Figure 5 | In vitro slice recording data related to Fig. 4,
and bassoon immunolabelling to determine monosynaptic connectivity
between CRF+ and SOM+ cells. a, Top, a light-evoked action potential
recorded in cell-attached mode from an example ChR2-expressing CRF+
cell. The average latency to light-evoked spike was 3.3 ±​ 2.2 ms (n =​  4).
Bottom left, ChR2 expression in the example CRF+ cell. Bottom right,
recorded neurons were filled with biocytin after recording. b, Whole-
cell recordings from CEl target cells were performed while stimulating
ChR2-expressing CRF+ cells with blue light. c, Top left, example histology
showing ChR2–eYFP fibres from CRF+ cells and biocytin-filled cells from
whole-cell patch-clamp recording. Top right, IPSCs evoked in a PKCδ​+
neuron by brief photostimulation of the CRF+ network (individual traces
from one cell in grey, corresponding average IPSC in black). Bottom,
immunohistochemistry performed after electrophysiology recordings
revealed PKCδ​ expression. d, Average IPSC amplitudes normalized to
cellular capacitance as approximation for cell size. e, Average latency
to IPSC peak from start of photostimulation. f, Average rise time of
IPSCs (10–90%). For d–f, no significant differences were observed
between SOM+ (n =​  23), PKCδ​+ (n =​ 8) or unidentified (n =​  13) cells
(Kruskal–Wallis test). g, Top, a light-evoked action potential recorded
from a ChR2-expressing SOM+ cell. The average latency to light-
evoked spike was 3.0 ±​ 1.1 ms (n =​ 3). Bottom left, ChR2 expression
in the example SOM+ cell. Bottom right, recorded neurons were filled
with biocytin after recording. h, Whole-cell recordings from CEl target
cells were performed while stimulating ChR2-expressing SOM+ cells
with blue light. i, Top left, example histology showing ChR2–eYFP
fibres from SOM+ cells and biocytin-filled cells from whole-cell patch-
clamp recordings. Top right, IPSCs evoked in a PKCδ​+ cell by brief
photostimulation of the SOM+ network. Bottom, immunohistochemistry
performed after electrophysiology recordings revealed PKCδ​ expression.
j, Average IPSC amplitudes normalized to cellular capacitance. No
significant differences were observed between CRF+ (n =​  24), PKCδ​
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTER RESEARCH
+
(n =​ 7) or unidentified (n =​ 8) cells (Kruskal–Wallis test). k, Average
latency to IPSC peak from start of photostimulation (Kruskal–Wallis test
P <​ 0.05; Dunn’s multiple comparisons test, no significant differences).
l, Average rise time of IPSCs (10–90%). No significant differences were
observed (Kruskal–Wallis test). m, Whole-cell recordings from CEl
SOM+ cells were performed while stimulating ChR2-expressing CRF+
cells with blue light. n, Top, IPSCs evoked in a SOM+ neuron by brief
photostimulation of the CRF+ network (individual traces from one cell in
grey, corresponding average IPSC in black). Bottom, IPSCs were blocked
by application of the GABAA-receptor antagonist picrotoxin. o, Whole-cell
recordings from CEl CRF+ cells were performed while stimulating ChR2-
expressing SOM+ cells with blue light. p, Top, IPSCs evoked in a CRF+
neuron by brief photostimulation of the SOM+ network (individual traces
from one cell in grey, corresponding average IPSC in black). Bottom,
IPSCs were blocked by application of the GABAA-receptor antagonist
picrotoxin. q, The amplitude of light-evoked IPSCs was significantly
reduced by application of picrotoxin for both
CRF to SOM (n =​ 3; reduction to 3.9% of baseline amplitude) and SOM
to CRF (n =​ 4; 2.8% of baseline) connectivity (two-tailed paired t-test).
r, Left, mCherry label of a SOM+ cell which received inhibitory input from
the CRF+ network. Middle, CRF–ChR2+ fibres come in close apposition
(arrowheads) to the biocytin-filled SOM+ cell (maximum intensity
projection). Right, single confocal planes of appositions labelled with
arrowheads in middle panel. Co-immunolabelling with the presynaptic
marker bassoon suggests monosynaptic connections between CRF+ and
SOM+ neurons. s, Left, tdTomato labelling of a CRF+ cell which received
inhibitory inputs from the SOM+ network. Middle, SOM–ChR2+ fibres
come in close apposition (arrowheads) to the CRF+ cell. Right, single
confocal planes of appositions labelled with arrowheads in middle panel.
Co-immunolabelling with the presynaptic marker bassoon suggests
monosynaptic connections between SOM+ and CRF+ neurons. Horizontal
lines throughout denote the mean. *​*​*​P <​  0.001.
 RESEARCH LETTER

a b
Rabies
AAV-flex-RabiesG and CEl CEm
RabiesΔG-mCherry-EnvA
CEl

BLA
CEm

BLA Glycoprotein
CEA
CRF-Cre-TVA or BLA CEl
CEm
SOM-Cre-TVA mice

BLA
d CRF SOM
IC
PVT c
BLA
vHP
vHIPP
BLP BLP
Apir PVT

40 20 0 20 40
% Input cells/total presynaptic cells APir

e
Target Target Number Number of Number of presynaptic cells
site cell type of mice starter cells Apir vHP BLA BLP PVT IC

375 812.3 628.7 645.3 560.7 335.3 663.7

CEA SOM+ 3
±49.7 ±237.7 ±181.2 ±151.4 ±172.7 ±139.7 ±201.1

15.8 140.0 55.5 28.8 114.8 37.5 22.0

CEA CRF+ 4
±4.3 ±62.6 ±31.1 ±7.0 ±60.8 ±23.6 ±10.0

f g h
AAV-flex-synaptophysin-GFP
CEl PH

LH
PeF
CEm
DM

MTu VMH
BLA
CEA BLA

Arc
100µm
µm 200µm
CRF-IRES-Cre
mouse

i j k
dmPAG

dl/lPAG
dl/lPAG
VMH

vlPAG
vlPAG
Arc

100µm 200µm 100µm

Extended Data Figure 6 | See next page for caption.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Extended Data Figure 6 | Input structures of CRF+ and SOM+ neurons,
and axonal targets of CRF+ cells. a, Pseudotyped-rabies tracing approach
used to identify input areas to the CEA CRF+ or SOM+ population.
A Cre-conditional AAV encoding the rabies glycoprotein and a G-deleted
EnVA-pseudotyped rabies were injected into the CEA of mice. Mouse
brain figure reproduced with permission from ref. 31. b, Left, merged
image of glycoprotein and rabies expression localized to CRF+ cells of
the CEA. Arrowheads indicate starter cells, which co-express rabies
and glycoprotein. Inset, an example starter cell expressing both rabies
and glycoprotein. Top right, rabies expression in the CEA. Bottom
right, neurons expressing glycoprotein. c, Example regions providing
monosynaptic input to the CRF+ network. vHIPP, ventral hippocampus;
BLP, basolateral amygdaloid nucleus, posterior part; PVT, paraventricular
thalamic nucleus. d, Comparison of the number of presynaptic cells to
CRF+ and SOM+ cells, normalized to the total number of presynaptic cells
in the defined regions. While there are a similar number of presynaptic
neurons across brain regions projecting to SOM+ cells (Friedman test,
P =​ 0.32), there are significantly different numbers of presynaptic neurons
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTER RESEARCH
projecting to CRF+ cells (repeated-measures one-way ANOVA F =​  14.39,
P <​ 0.05). APir, amygdalopiriform transition area; BLA, basolateral
amygdala; IC, insular cortex. e, Table listing all raw values from rabies-
tracing experiments. f, Injection strategy used to locate terminal fields
of CEA CRF+ axonal projections. A Cre-conditional AAV encoding
synaptophysin conjugated to GFP was injected into the CEA of CRF-ires-
cre mice. Mouse brain figure reproduced with permission from ref. 31.
g, Expression of GFP in the CEA illustrating transfected somata and GFP+
puncta representing local axon terminals. h, CRF+ neurons project to
multiple structures in the hypothalamus. PH, posterior hypothalamic
area; LH, lateral hypothalamic area; PeF, perifornical nucleus; DM,
dorsomedial hypothalamic nucleus; MTu, medial tuberal nucleus; VMH,
ventromedial hypothalamic nucleus; Arc, arcuate hypothalamic nucleus.
i, Demonstration of CRF+ terminals within the VMH. j, CEA CRF+
neurons also project to the peri-aqueductal grey (PAG). dmPAG,
dorsomedial PAG; dl/lPAG, dorsolateral/lateral PAG; vlPAG, ventrolateral
PAG. k, CEA CRF+ cells project widely throughout PAG columns.