R ES E A RC H

◥ We aligned spiking activity with epochs of push-

RESEARCH ARTICLE ing (including push initiation and push-back),
resistance, retreat, and stillness (Fig. 1H and meth-
ods). The average firing rate of recorded pPyr
NEUROSCIENCE units was significantly higher during the “effort-
ful” (push and resistance) behavioral epochs, but

History of winning remodels not the passive (retreat) behavioral epochs, than
during stillness (Fig. 1, H and I, and movie S2). In
contrast, the average firing rate of pIN units in-
thalamo-PFC circuit to reinforce creased insignificantly during the retreat epoch
(Fig. 1J). A notable fraction of the pPyr units showed

social dominance increased firing rates during push (11.4%, 31 of

271 units) and resistance (10.1%, 25 of 247 units)
behaviors, significantly more than the fraction that
Tingting Zhou,1,2,3,4* Hong Zhu,1,2,3,4* Zhengxiao Fan,3,4 Fei Wang,1,2 Yang Chen,1 showed decreased firing rates (1.1% for pushing, P <
Hexing Liang,3,4 Zhongfei Yang,1 Lu Zhang,5 Longnian Lin,6 Yang Zhan,7 0.0001, Z test; 2% for resistance, P = 0.002, Z test;
Zheng Wang,1 Hailan Hu3,4,8† Fig. 1K and fig. S2E). One-third of neurons (11 of 31)
with increased activity during push behavior also
Mental strength and history of winning play an important role in the determination of social showed an increase in firing rate during resistance,
dominance. However, the neural circuits mediating these intrinsic and extrinsic factors which was significantly higher than the chance
have remained unclear. Working in mice, we identified a dorsomedial prefrontal cortex level (P < 0.0001, Z test; Fig. 1L). These 11 neurons
(dmPFC) neural population showing “effort”-related firing during moment-to-moment showed highly correlated increases in activity dur-
competition in the dominance tube test. Activation or inhibition of the dmPFC induces ing push and resistance behaviors [linear regres-
instant winning or losing, respectively. In vivo optogenetic-based long-term potentiation sion analysis, coefficient of determination (R2) =
and depression experiments establish that the mediodorsal thalamic input to the dmPFC 0.82, slope = 0.90, P < 0.001; Fig. 1M]. Although
mediates long-lasting changes in the social dominance status that are affected by history both pushing and resistance require effort, they
of winning. The same neural circuit also underlies transfer of dominance between different differ in that the former involves body movement,
social contests. These results provide a framework for understanding the circuit basis of whereas the latter does not. Thus, regardless of
adaptive and pathological social behaviors. whether the animal is in motion, these two dif-
ferent behavioral states tend to recruit the same

I
n most social species, reaching the top of the
social hierarchy is strongly affected by mental
strength or personality traits (including cour-
age, perseverance, and motivational drive), as
well as a previous history of winning (1–6).
The formation of a dominance hierarchy can
result from a reinforcing mechanism known as
the “winner effect,” where animals increase their
probability of victory after prior winning (7–11).
The dorsomedial prefrontal cortex (dmPFC) has
been implicated in the chronic regulation of social
dominance (12–18). However, the acute require-
ment of the dmPFC during ongoing social com-
petition and the upstream neural circuits that
regulate dmPFC activity in dominance behav-
iors have been essentially unknown. It has also
been unclear whether the winner effect can be
generalized—in other words, whether dominance
acquired in one type of competition can transfer
to another behavioral type.
1
Institute of Neuroscience and State Key Laboratory of
Neuroscience, Shanghai Institutes for Biological Sciences,
Chinese Academy of Sciences, Shanghai 200031, P.R. China.
2
Graduate School of Chinese Academy of Sciences, Shanghai
200031, P.R. China. 3Interdisciplinary Institute of Neuroscience
and Technology, Qiushi Academy for Advanced Studies,
Zhejiang University, Hangzhou 310012, P.R. China. 4Center
for Neuroscience, Key Laboratory of Medical Neurobiology
of the Ministry of Health of China, School of Medicine, Zhejiang
University, Hangzhou 310058, P.R. China. 5Shanghai Center
for Mathematical Sciences, Fudan University, Shanghai 200433,
P.R. China. 6Key Laboratory of Brain Functional Genomics, East
China Normal University, Shanghai 200062, P.R. China. 7Brain
Cognition and Brain Disease Institute, Shenzhen Institute of
Advanced Technology, Chinese Academy of Sciences, Shenzhen
518055, P.R. China. 8Mental Health Center, School of Medicine,
Zhejiang University, Hangzhou 310013, P.R. China.
*These authors contributed equally to this work.
†Corresponding author. Email: huhailan@zju.edu.cn
Winner mice display more pushes and
resistances in the dominance tube test
To investigate social competition in laboratory
mice, we applied the dominance tube test (19),
which is highly transitive and stable and cor-
relates well with other dominance measures
(6). The behavior in the tube test was video-
monitored and categorized as push initiation,
push-back, resistance, retreat, or stillness (Fig. 1,
A and B; fig. S1A; movie S1; and methods).
Analysis of 72 tube test trials revealed that win-
ner mice initiated significantly more pushes, and
with a longer duration per push, than loser mice
(Fig. 1C). When being pushed, winner mice also
showed more and longer push-backs and resist-
ances and fewer retreats (Fig. 1, D to F). For a
cage of four male weight-matched C57BL/6J
mice, we derived a linear dominance rank order
based on total numbers of wins against cagemates
in pairwise tube tests (18) (fig. S1B). Opponents with
closer rank distances spent a longer time and gen-
erated more pushes (fig. S1, C and D) in the tube.
dmPFC neurons activated in effortful
behavioral epochs during social contests
We performed single-unit recordings in freely
behaving mice while they engaged in the tube
test (Fig. 1G). Using 16-channel tetrodes targeting
dmPFC [including the anterior part of the anterior
cingulate (ACC) and the prelimbic (PL) part of the
PFC (fig. S2, A and B)], we successfully recorded
342 well-isolated neurons in 22 mice, including
306 putative pyramidal (pPyr, 89.5%) and 36
putative fast-spiking interneuron (pIN) cells (fig.
S2, C and D; criteria for single-unit isolation and
cell-type classification are given in the methods).
subset of dmPFC neurons.
DREADD inhibition of dmPFC reduces
effortful behaviors and causes losing
Given that dmPFC neuronal activity is differen-
tially modulated when animals make an effort dur-
ing competition, we next tested whether dmPFC
neural activity is required for dominance in the
tube test. We applied the DREADD (designer re-
ceptors exclusively activated by designer drugs)
method to inactivate dmPFC neurons, using an
adeno-associated virus (AAV2) expressing the engi-
neered Gi-coupled hM4D receptor (20) (Fig. 2A
and fig. S3, A and B). Whole-cell recordings of
neurons from acutely isolated dmPFC brain slices
confirmed that clozapine-N-oxide (CNO, 5 mM)
suppressed hM4D-expressing dmPFC neuron activ-
ity (Fig. 2B), causing a significantly increased
spike threshold and decreased spike number
under current step injections (Fig. 2, C and D). We
then intraperitoneally (i.p.) injected CNO (5 mg
per kilogram of body weight) into a subset of mice
(each from a different four-mouse group), which
had hM4D expressed bilaterally in the dmPFC
and had stable tube test ranks (persisting for at
least three continuous daily trials before the ma-
nipulation; Fig. 2E). This treatment induced a de-
cline in tube test ranks of the injected mice, starting
at 1 to 1.5 hours and peaking at 6 to 8 hours after
injection (Fig. 2, E to G and fig. S3C). There were
significantly fewer and shorter initiated pushes
and push-backs, and more retreats, after CNO
injection (Fig. 2, H and I). At 24 hours after CNO
injection, most mice returned to their original
rank position (Fig. 2, E to G), consistent with the
reversal of the effect on cell physiology after CNO
washout (Fig. 2B).

Zhou et al., Science 357, 162–168 (2017) 14 July 2017 1 of 7

R ES E A RC H | R E S EA R C H A R T I C LE

Optogenetic activation of dmPFC CAG promoter (AAV-CAG-ChR2-tdTomato) was pulses per second; fig. S5A) with 473-nm laser
induces instantaneous winning in the stereotactically injected into the right dmPFC of stimulation, which significantly increased the
tube test the ranked mice and expressed for 4 weeks, expression of the immediate early gene c-Fos in
Using optogenetics, we next tested whether dmPFC and an optic fiber was implanted directly above the illuminated side of the dmPFC (P < 0.05;
activation is sufficient to quickly induce domi- the injection site (Fig. 3, A and B; fig. S4; and Fig. 3B). Whole-cell recordings in dmPFC brain
nance behavior in social competition. AAV2 virus methods). dmPFC neurons show both phasic and slices showed that both a 100-Hz phasic protocol
expressing light-sensitive channelrhodopsin (ChR2) tonic firing patterns (22). We thus first used a and a 5-Hz tonic protocol (fig. S5B) efficiently
(21) under the control of the ubiquitously expressed 100-Hz phasic protocol (9.9 ms per pulse, four activated dmPFC neurons (Fig. 3, C and D), and

Push-initiated Push-back Resistance Retreat

Winner Loser Winner Loser Winner Loser Winner Loser
Stillness

Number of behaviors /trial

Number of behaviors /trial

Push-back **** **** **** 4 **** **** ****

% time retreating while

2.0 3 2.0 100 80

Mean duration /push (s)

Mean duration /push (s)

% time resisting while

Push-initiated Resistance

being puhsed
80

being pushed
1.5 1.5 3 60
Retreat 2
60
1.0 1.0 2 40
Winner 40
1
Loser 0.5 0.5 1 20
20
0 1 2 3 4 5 72 72 68 23 31 69 21 17 31 71 31 71
Time from meeting (s) 0.0 0 0.0 0 0 0

Trial Push Resistance Retreat Stillness n.s.

***

Mean FR of all pPyr units (Hz)

9

Mean FR of all pIN units (Hz)

5
5 **
*** ***
4
15 * **
3 **

Mean FR (Hz)
4 6
10
2
5
PC2 1
0
0 5 10 15 20 3 3
PC1 Time from meet (s)

Push Resistance
Push Resistance Retreat increase 10 increase
n.s.
4 **** 4 *** 19 10
Mean FR (Hz)

4
Mean FR (Hz)
Mean FR during stillness (Hz)

Mean FR during stillness (Hz)

Mean FR during stillness (Hz)

Mean FR (Hz)

1
25 25 25 1 4
Retreat
3 20 3 20 3 increase
20
5
2 11.4% 15 2 10.1% 15 2 6.8% 0.8
15 p < 0.001
1.1% 2.0% 4.9%
R2 = 0.82
Resistance CI

10 10 0.6
10
5 5 0.4
5 87.5 % 87.9 % 88.3 %

0 0 0.2
0 0 5 10 15 20 25
0 5 10 15 20 25 0 5 10 15 20 25
Mean FR during retreat (Hz) 0.0
Mean FR during push (Hz) Mean FR during resistance (Hz) 0.0 0.2 0.4 0.6 0.8
Increase Decrease No change Push CI

Fig. 1. dmPFC neurons are activated during effortful behaviors in the behavioral epochs. One-way repeated measures ANOVA (analysis of variance).
dominance tube test. (A) Schematic of “effortful” (push initiation, push- (K) Scatter plots of the firing rates of all pPyr units during push (left),
back, and resistance) and passive (stillness and retreat) behavior patterns of resistance (middle), and retreat (right) behavioral epochs, plotted against firing
two mice confronting each other in the tube test. The arrows indicate the rates during the stillness epoch. Colored circles indicate neurons that showed
direction of body movement. (B) Sample behavior annotations for a pair of significant differences in firing rates. Most of the circles are distributed
mice in a tube test trial. (C and D) Number and mean duration of pushes underneath the y = x diagonal line in the push and resistance plots, but not in
initiated (C) and push-backs (D). The number of trials is indicated in each bar. the retreat plot. Pie graphs show the percentage of pPyr neurons that had
Mann-Whitney U test. (E and F) Percentage of time that mice resisted (E) or significantly higher, significantly lower, or unchanged firing rates during the
retreated (F) while being pushed. Mann-Whitney U test. (G) Tetrode recording respective epochs, relative to stillness. Wilcoxon signed-rank test. (L) Venn
in the dmPFC of mice during the tube test (top) and three well-isolated single diagram of overlap between subpopulations with increased activity during
units (yellow, red, and blue clusters; bottom). PC, principal component. push, resistance, and retreat behaviors. (M) Linear regression analysis of the
(H) Sample raster plots of a pPyr neuron during five losing tube test trials. change index (CI; see methods) of push and resistance behaviors for the
Different behavioral epochs are indicated by colored shading. Inset, mean firing 11 neurons that showed significantly increased activity during both states.
rate (FR) for different behavioral epochs. Kruskal-Wallis test. (I and J) Mean Dashed lines indicate 95% confidence intervals. *P < 0.05; **P < 0.01; ***P <
firing rate of all pPyr (n = 160) (I) and pIN (n = 13) (J) units during different 0.001; ****P < 0.0001; n.s., not significant. Error bars, SEM.

Zhou et al., Science 357, 162–168 (2017) 14 July 2017 2 of 7

R ES E A RC H | R E S EA R C H A R T I C LE

this was also verified by in vivo optrode record-
ings of the single-unit responses from virally
injected animals (Fig. 3, E and F).
In tube tests with pairs of mice, we delivered
the 100-Hz phasic light protocol to one of the
mice to activate its dmPFC immediately before
it entered the tube to confront its opponent, and
we kept the light on throughout the test, which
lasted, on average, 12.8 ± 1.3 s (n = 93; Fig. 3G).
This instantaneously induced winning against
previously dominant opponents with a 90% suc-
cess rate (Fig. 3, G and H; fig. S5C, J and K; and
movie S3), without affecting the motor perform-
ance or anxiety level (fig. S6). The same photo-
stimulation protocol did not affect the tube test
rank of mice injected with AAV-Ubi-eGFP virus
(eGFP, enhanced green fluorescent protein) in
the dmPFC (Fig. 3I and fig. S5F). The 5-Hz tonic
stimulation protocol had a similar success rate
(80%) in elevating the rank of stimulated mice
(Fig. 3J and fig. S5, D, J, and K).
Under photostimulation, the originally sub-
ordinate mice not only resisted pushes from the
opponents for a longer duration, but also pushed
more (Fig. 3K). The laser intensity required to
dominate the opponents correlated with the rank
distance that the mouse needed to move: A rank
increase of three (rank 4 moving to rank 1) re-
quired laser intensity that was 3.2-fold that re-
quired for a rank increase of two (rank 4 to rank
2 or rank 3 to rank 1) and 7.1-fold that required
for a rank increase of one (rank 4 to rank 3, rank
3 to rank 2, or rank 2 to rank 1), demonstrating a
dosage-dependent relationship between the level
of dmPFC activation and the amount of effort
required (Fig. 3L).
Importantly, dmPFC photoactivation did not
change muscle strength, as assessed by measuring
grip strength during the light-on and -off periods
(fig. S7A). Most (98%, n = 103) photostimulated
tube test wins occurred within less than 1 min,
suggesting that testosterone, a conventionally
slow-acting hormone, is unlikely to have played
a role in this quick process. To confirm this, we
measured the levels of testosterone at 1 min and

Fig. 2. DREADD inhibition of dmPFC neurons causes losing and
decreases effortful behaviors in the tube test. (A) AAV-hSyn-hM4D
construct and viral injection site in the dmPFC, including the PL region and
part of the ACC. Scale bar, 50 mm. IRES, internal ribosome entry site; WPRE,
woodchuck hepatitis virus posttranscriptional regulatory element; pA, poly(A);
MO, medial orbital cortex. (B) Current-voltage relationship of a representative
dmPFC neuron recorded before, during, and after 5 mM CNO perfusion. Raw
traces show individual voltage responses to a series of 600-ms current pulses
from 0 to 120 pA with 20-pA steps. Red traces indicate the minimal current to
induce action potentials. (C) The minimal injected current to induce action
potential (APs) is increased by CNO. Paired t test, n = 10. (D) Number of induced
action potentials at different current steps. Paired t test, n = 10. (E) Tube test
results for a cage of hM4D-expressing mice before and after i.p. injection of
CNO into the rank-1 mouse at time 0. hr, hour. (F) Summary of rank changes in
hM4D-expressing mice after i.p. injection of CNO. Each line represents one
animal. (G) Average rank change after CNO or saline injection. Wilcoxon
matched-pairs signed-rank test. (H) Behavioral annotation of two tube test trials
between the same pair of mice before and after CNO injection. (I) Comparison
of behavioral performance of same mice injected with saline solution or CNO.
Mann-Whitney U-test. *P < 0.05; **P < 0.01; ***P < 0.001. Error bars, SEM.

Zhou et al., Science 357, 162–168 (2017) 14 July 2017 3 of 7

R ES E A RC H | R E S EA R C H A R T I C LE

1.5 hours after photostimulation and found them gression levels and no significant difference anterior part of the ACC and the PL region
to be indistinguishable from the baseline and un- between the light-on and -off periods in aggres- (Figs. 2A and 3A and fig. S4). A more dorsal and
stimulated control levels (fig. S7B). To test whether sive behaviors (including attacks and chases; posterior part of the ACC (Fig. 3M and fig. S8, A
the increased dominance is manifested by a height- fig. S7, C to F, and movie S4). Furthermore, the and B) and a ventral mPFC site mostly contain-
ened level of basal aggression, we subjected mice dmPFC-photostimulated mice also showed a nor- ing the infralimbic cortex (Fig. 3M and fig. S8,
to the resident-intruder assay (23), during which mal preference toward novel mice in the social C and D) were not effective in stimulating winning
we intermittently turned the 473-nm light (5 Hz, memory test (fig. S7, G and H). behavior (Fig. 3M and fig. S5, G and H). For cell-
10 ms) on and off in 1-min epochs with a switched With localized injection, we mapped the effec- type specificity, we used a CaMKII (calcium- and
sequence. The test mice showed low basal ag- tive site to a dmPFC site containing both the calmodulin-dependent protein kinase II) promoter

In vitro whole-cell recording In vivo optrode recording In vivo optrode recording

CAG ChR2-TdTomato WPRE pA
Phasic 100 Hz Phasic Tonic
100 Hz 5 Hz
ACC

PL 50
40 60

Spikes

Spikes
30 40
MO Bregma 20 20
+2.10 mm 10
0 0
-50 0 50 100 -40 -20 0 20 40 60
ChR2 / c-Fos / NeuN
80 * Tonic 5 Hz 1 1
Uninjected side Injected side
c-fos cells #

Cell No.

Cell No.
40

Z-Score

Z-Score
0
32 32
ted ted
jec jec -50 0 50 100 -200 -100 0 100 200
U nin In Time (ms) Time (ms)

CAG::ChR2 Phasic 100 Hz; n = 10

5 Hz; n = 10 CaMKII::ChR2; n = 8
3 2 1 Hz; n = 7 eGFP, 5 Hz; n = 6
Stimulated mouse Ubi::eGFP 2
Cagemates n = 10
Phasic
Rank change
Rank change

Rank change
2 100 Hz n=6 **
** * *
Tube test rank

1 1
1 **
2 1 *
3 * * *
4 0 0
-3 -2 -1 0
-3 -2 -1 0 0
-3 -2 -1 0 1 2 3 (d) 1 2 3 (d) 1 2 (d) -3 -2 -1 0 1 2 3 (d)

Light off Light on, rank changed dmPFC, n = 10

Neocortex 2
* IL, n = 8
Normalized laser intensity

**** ** *** *** 12 anterior posterior

8 2 100 100 pACC, n = 8
Mean duration /push (s)

Rank change

dorsal
% time retreating
% time resisting
# of pushes /trial

6 80 80 8 ACC 1 **
60 60
4 1
PL * *
40 40 4 IL
*
2 dmPFC 0
20 20 ventral
0 10 10 3
72 18 72 18 41 16 41 16 pACC
0 0 0 0 1 2 3
Pushes Pushes Resistance Retreat Rank distance IL -3 -2 -1 0 1 2 3 (d)

Fig. 3. Optogenetic activation of dmPFC Pyr neurons induces instant photostimulation of the rank-3 mouse at day 0. (H and I) Summary
winning and more effortful behaviors in the tube test. (A) Schematic of dmPFC photostimulation–induced rank change in mice injected
illustrating the CAG::ChR2 viral construct, viral injection site, and optic with CAG::ChR2 (H) or Ubi::eGFP (I) virus. Each line represents one
fiber placement (indicated by the white arrowhead). Scale bar, 100 mm. animal. (J) Rank change in the tube test under different stimulation
(B) c-Fos expression induced by photostimulation (NeuN, neuron nuclei). conditions. Light stimulation was delivered throughout the tube test at
Scale bar, 50 mm. Student’s t test. (C and D) Current-clamp traces from day 0. Wilcoxon matched-pairs signed-rank test. Except as noted,
an in vitro slice recording of a ChR2-expressing neuron illuminated by CAG-ChR2 mice were used. (K) Comparison of the behavioral perform-
100-Hz phasic (C) or 5-Hz tonic (D) light stimulation. Scale bars in (C), ances of same mice during light-off and light-on trials. Mann-Whitney
100 ms (horizontal) and 20 mV (vertical); in (D), 1 s and 40 mV. (E and U test. (L) A greater rise in rank position requires a stronger laser
F) Raster plots (top), peristimulus time histogram (middle), and waveform intensity. Wilcoxon signed-rank test. (M) Left, schematic illustrating
(inset; scale bars, 500 ms and 50 mV) from an in vivo optrode recording mPFC subregions. The dmPFC contains both anterior ACC and the PL
of a single neuron responding to 100-Hz phasic (E) and 5-Hz tonic (F) region. IL, infralimbic; pACC, posterior ACC. Right, rank change in the
light stimulation. The Z-scored single-unit firing rates of multiple neurons tube test after optogenetic activation of different mPFC subregions.
are shown in the bottom panels. (G) Daily tube test results for a cage of Wilcoxon matched-pairs signed-rank test. *P < 0.05; **P < 0.01; ***P <
mice injected with CAG::ChR2 virus before and after acute dmPFC 0.001. Error bars, SEM.

Zhou et al., Science 357, 162–168 (2017) 14 July 2017 4 of 7

R ES E A RC H | R E S EA R C H A R T I C LE

to drive ChR2 only in dmPFC Pyr neurons, and we

1d
Maintain new rank *** Saline; n=5 MK801; n=6 found that it was sufficient to elevate tube test rank
100 **

% mice maintain new rank

Return to original rank (Fig. 3J and fig. S5, I to K).

% stimulated mice
100
80 MK801/saline 6X wins
9
2
* 80 Synaptic strength in the MDT-dmPFC
Rank change

60 11
n = 15 circuit underlies the winner effect

Rank change
2 60
40 in the tube test
1 40
20 4 1 n.s. Performance in the tube test diverged on the
n=9 20
0 0 second day after photostimulation: Some mice
2-5 6-20 0 0 returned to their original rank, whereas others
Number of e 01
-3 -2 -1 0 1 2 3 (d) lin maintained their newly elevated rank (Fig. 3H
stimulation trials -3 -2 -1 0 1 2 3 (d) Sa MK8
and 4A). Comparison of their experiences revealed
baseline that these mice differed in the number of winning
hSyn oCHIEF-Tdtomato WPRE pA
after 6X wins trials on the stimulation day: Mice receiving more
500uV
dmPFC than six photostimulated wins all maintained
fEPSP 10ms
their new rank, whereas most mice receiving
1.6 ctrl; n=5 6X wins; n=8 fewer than five photostimulated wins returned

fEPSP (normalized)
Bregma +2.43 mm
6X wins ** to their original rank (Fig. 4A). This sustained
dmPFC
1.4 * rank elevation was not caused by a secondary
LHb MHb
1.2
effect through the experience of nonstimulated
MD MDL rival mice (fig. S9). Thus, repeated stimulated
MD
1.0 winning led to sustained dominance without
MDC
MDM further photostimulation, reflecting the winner
0.8
Bregma +1.46 mm -2 -1 0 1 2 3 (d) effect. Systematic injection of MK801 (0.15 mg/
kg, i.p.) eliminated the sustained winning in-
6X wins + LFS; n=6 0.5 hr duced by six photostimulated wins, suggesting
baseline

% mice maintain new rank

100uV post LFS 6X wins; n=7 100 ** that an NMDAR (N-methyl-D-aspartate receptor)–
10ms 6X wins LFS dependent plasticity mechanism may mediate
80
fEPSP (normalized)

1.2 LFS this long-lasting increase in dominance (Fig. 4B

2 * 60 and fig. S10A).
Rank change

1.0
40 We next searched for the neural circuit path-
0.8 1 n.s. 20
way that supports this behavioral plasticity mecha-
nism. The mPFC receives prominent projections
0.6 0 0 from the mediodorsal thalamus (MDT) (24), and
l
-72 -48 -24 0 0.5 24 48 72 (h) Ctr LFS the MDT-dmPFC circuit shows synaptic weaken-
-30 0 30 60 (min)
ing during repeated defeat–induced social avoidance
(25). We thus hypothesized that this same pathway
% of mice show sustained win

baseline
oCHIEF, HFS in homecage; n=8 72 hr
may undergo long-lasting synaptic strengthening
post HFS
GFP, HFS in homecage; n=6
60 * after repeated winning. If that is the case, we
20uV HFS
fEPSP (normalized)

2
2.0 10ms * 40
should be able to (i) detect enhanced synaptic
Rank change

HFS strength in the MDT-dmPFC pathway after re-

1.5 1 peated winning, (ii) eliminate the sustained win-
20 ning by introducing long-term depression (LTD)
1.0
0 0
to reverse the synaptic strengthening in the MDT-
0.5 mPFC circuit, and (iii) directly cause sustained
rl
-30 0 30 60 (min) -72 -48 -24 0 3 24 48 72(h) Ct HFS winning by inducing long-term potentiation (LTP)
in the MDT-mPFC synapses without any tube
Fig. 4. Synaptic strength of the MDT-dmPFC circuit underlies the winner effect in the tube test competition. We expressed oCHIEF, a var-
test. (A) Mice either maintain their new rank or return to their original rank position after dmPFC iant of ChR2 that can faithfully respond to 100-Hz
photostimulation, depending on the number of stimulated-win trials. Z test, n = 13 and 11 mice that stimulation (fig. S11A) (26, 27), in the MDT (fig.
underwent two to five and six to 20 stimulation trials, respectively. d, day. (B) I.p. injection of MK801, S12, A to C) and implanted an optrode (for elec-
but not saline solution, at 30 min before six photostimulated wins abolished sustained winning on trophysiology recording) or an optic fiber (for
subsequent days. Left, Wilcoxon signed-rank test; right, Z test. (C) Optogenetic LTP or LTD experiment in behavioral manipulation) in the dmPFC (Fig. 4C;
the MDT-dmPFC pathway. Top, schematic of viral injection and opotrode recording sites. Bottom left, fig. S12, D and E; and methods). Optical activa-
representative coronal section showing the injection site of AAV-oCHIEF-tdTomato in MDT. Bottom right, tion of the MDT axonal terminals in the dmPFC
representative coronal dmPFC section showing distribution of tdTomato+ axons projected from the (5 Hz, 10 ms) induced instantaneous winning in
MDT. LHb, lateral habenular; MHb, medial habenular; MDC, mediodorsal thalamus, central; MDM, the tube test (fig. S12F).
mediodorsal thalamus, medial; MDL, mediodorsal thalamus, lateral. Blue, Hoechst; red, tdTomato. To measure synaptic responses of the MDT-
Scale bars, 25 mm (top right) and 500 mm (bottom right). (D) Average slopes of in vivo light-evoked dmPFC pathway in free-moving mice, we recorded
fEPSPs (average of six responses) in the dmPFC (normalized to baseline) before and after six wins field response in the dmPFC evoked by photo-
induced by photostimulation of MDT-dmPFC terminals. Insets, representative fEPSP trace before and stimulation of the oCHIEF-expressing MDT-
after six wins. Paired t test. (E and G) Average slopes of in vivo light-evoked fEPSPs (average of six dmPFC axonal terminals. Brief pulses of blue
responses) in the dmPFC (normalized to baseline) before and after LFS (E) or HFS (G). Insets, light were delivered at 0.05 Hz through the opti-
representative fEPSP traces before and after LFS (E) and HFS (G). (F) LFS reverses the sustained rank cal fiber, and fEPSPs (field excitatory postsynaptic
increase resulting from six wins induced by MDT-dmPFC photostimulation. Left, Wilcoxon signed-rank potentials) were used to measure the MDT-dmPFC
test; right, Z test. (H) HFS induction directly causes sustained winning in the tube test. Left, Wilcoxon synaptic efficacy (fig. S11, B and C) (27). After a
signed-rank test; right, Z test. *P < 0.05; **P < 0.01; ***P < 0.001. Error bars, SEM. stable baseline was acquired for at least 3 days,

Zhou et al., Science 357, 162–168 (2017) 14 July 2017 5 of 7

R ES E A RC H | R E S EA R C H A R T I C LE

800 Rank1 in tube test

Cold floor

Time (s)

Rank in warm spot test

Rank2 in tube test 600

Time in warm spot (s)

Time in warm spot (s)

4
Rank3 in tube test 3 1 1 4
Warm spot 800 400
Rank4 in tube test
0 400 0 1 4 4

3
600 1 2 3 4
Tube test rank
400
0 5 4 0

2
200
200 R2 = 0.91
n=9 6 2 0 1

1
0 0
0 5 10 15 20 (min) 1 2 3 4 1 2 3 4
Total time Rank in tube test Rank in tube test
CNO Saline Before 10X winning trials Win without light After
n.s. n.s.
*
Time in warm spot (s)

Time in warm spot (s)

1500 1500
n.s. n.s.
* * * * *
1000 1000 *

500 500 No rank elevation

Warm spot test in tube test
Winning in
Sustained rank elevation
0 0 * * contest type A

Rank in warm spot test

Time in warm spot (s)

n.s. n.s. 1000 in tube test

* * n.s. n.s. 1 2 **

in warm spot test

800
warm spot test

warm spot test

1 1

Rank change
600 2 Dominance
Rank in

Rank in

2 2
400 3 1
3 3
200 4 16 6 Winning in
4 4
-1 d 2 h 2 d -1 d 2 h 2 d 0 contest type B
Before After Before After 0
Time from injection Time from injection

Fig. 5. Transferrable dominance from the tube test to the warm spot the experimental procedure. After basal ranks were assessed in the tube
competition. (A) Schematic of the warm spot test. Four mice compete test and warm spot test, dmPFC photostimulation was applied to
for a warm corner in a cage with an ice-cold floor. (B) Cumulative time subordinate mice during the tube test to induce winning in 10 successive
in the warm spot of four cagemate mice of different tube test rank. trials; mice were confirmed to maintain the elevated tube test rank in
Inset, total time in the warm spot in the 20-min test. (C) Correlation the absence of photostimulation and then subjected to the warm spot
between time in the warm spot and tube test rank. Pearson’s correlation test. The red asterisk marks the mouse being manipulated. (H) Time
test, P = 0.046. (D) Contingency table showing the correlation between in the warm spot (left) and rank (right) in the warm spot test were
rank in the tube test and rank in the warm spot test. The number of elevated 2 hours after repeated winning in the tube test. Wilcoxon
animals in each category is displayed. (E and F) Time in the warm spot matched-pairs signed rank test, n = 6. (I) Rank increase in the warm spot
(top) and rank in the warm spot test (bottom) for mice expressing hM3D test is specific to mice with a sustained rank increase in the tube test.
in the dmPFC at 1 day before, 2 hours after, and 2 days after i.p. injection Mann-Whitney U test. (J) Reciprocal reinforcement of winning in different
of CNO (E) or saline solution (F). Data for each mouse are shown in contests leads to establishment of dominance hierarchy. *P < 0.05; **P <
gray; averages are shown in red and blue; n = 6. (G) Schematic of 0.01. Error bars, SEM.

we photostimulated the MDT-dmPFC axonal Transferrable winner effect from the stimulated tube test wins, their occupation times
terminals to induce repeated winning six times tube test to the warm spot competition and ranks in the warm spot test were significantly
and measured the fEPSPs on the following days. We next asked whether dominance acquired in elevated compared with their performances be-
Accompanying the rank elevation, the fEPSP of the tube tests can transfer to other forms of domi- fore the tube test (P = 0.016 for occupation time,
the MDT-dmPFC pathway was significantly in- nance behavior. We subjected mice to an inde- P = 0.016 for rank, Wilcoxon signed-rank test;
creased on the following 2 days after repeated pendent measure of social hierarchy, the warm Fig. 5, G to I), even though these mice were not
winning, reflecting strengthening of the MDT- spot test, in which four cagemate C57 mice com- directly stimulated in the warm spot context.
dmPFC synapses (Fig. 4D). peted for a warm corner in a cage with an ice-
Next, by applying low-frequency stimulation cold floor (Fig. 5A). The amount of time that each Discussion
(LFS), an optical LTD protocol (900 pulses of 2-ms mouse occupied the warm spot correlated with Collectively, these results provide strong evidence
light stimulation delivered at 1 Hz; fig. S11D) its tube test rank (Fig. 5, B to D, and movie S5), that activation of the dmPFC is both necessary
(27, 28), at the MDT-dmPFC terminals, we were cross-validating that these two assays share and sufficient to quickly induce winning in social
able to induce LTD in the MDT-dmPFC synapses dominance as a common core variable. Activation competitions. Specifically, by optogenetically isolat-
(Fig. 4E) and eliminate the sustained winning of dmPFC neurons by using AAV2 expressing ing a synaptic input from the MDT to the dmPFC,
effect after six stimulated wins (Fig. 4F and fig. the engineered Gq-coupled hM3D receptor (20) we could selectively manipulate synapses driven
S10B). Conversely, delivery of high-frequency increased the occupation time and rank in the by this input and establish a causal relationship
stimulation (HFS), an optical LTP protocol (fig. warm spot test at 2 hours, but not 2 days, after between the activity of the MDT-dmPFC circuit
S11A and methods) (28), at the MDT-dmPFC CNO injection (1 mg/kg, i.p.) (Fig. 5, E and F). and dominance behavior. dmPFC activation does
terminals significantly potentiated MDT-dmPFC We then tested whether sustained winning in not seem to boost dominance by enhancing basal
transmission in vivo (Fig. 4G). Application of this the tube test resulting from repeated dmPFC stim- aggression level or physical strength (fig. S7), but
in vivo LTP protocol to freely moving mice in the ulation could lead to increased rank in the warm rather by initiating and maintaining more effort-
home cage caused significant, long-lasting tube spot test. After we subjected mice that were pre- ful behaviors in social competition (Fig. 3K). The
test rank increases (Fig. 4H and fig. S10C). viously subordinate in both tests to 10 photo- mPFC has been implicated in cost-benefit analysis

Zhou et al., Science 357, 162–168 (2017) 14 July 2017 6 of 7

R ES E A RC H | R E S EA R C H A R T I C LE
and effort-based decision-making (14–16, 29–32).
We propose that these mPFC-based cognitive pro-
cesses may provide a neurobiological foundation
for dominance-associated personality traits, such
as perseverance or competitive drive.
One important parameter for the cost-benefit
computation in a social confrontation is the his-
tory of winning. With the in vivo optogenetic LTP
and LTD experiments, we provide evidence that
synapses in the MDT-dmPFC pathway may en-
code winning history. Whereas earlier work on
the winner effect in fish was mostly focused on
hormonal changes after repeated winning (33),
our results reveal that the synaptic plasticity
mechanism in the MDT-dmPFC circuit plays a
key role in the winner effect in mammals. More-
over, we discovered a generalized form of the
winner effect, where dominance transfers from
one contest type to another through a shared
neural circuit mechanism. Previous studies of
the winner effect were restricted to the impact of
winning on the same behavior paradigm (33).
However, given that animals are dealing with
different forms of competition in setting up the
social hierarchy, the generalized winner effect
that we describe here is of high evolutionary
importance—for example, it may allow a monkey
that succeeds in fighting for bananas earlier to
occupy a more comfortable resting spot later.
Such reciprocal reinforcement between winning
in different behavioral paradigms would help to
accelerate the establishment of a stable domi-
nance hierarchy (Fig. 5J). It may also have im-
portant implications in cognitive training for
competitive games. Considering that an excess
or lack of dominance drive is associated with many
personality disorders and mental problems, our
Zhou et al., Science 357, 162–168 (2017) 14 July 2017
results might shed light on the treatment of
these psychiatric diseases.
RE FERENCES AND NOTES
1. M. F. Rushworth, N. Kolling, J. Sallet, R. B. Mars, Curr. Opin.
Neurobiol. 22, 946–955 (2012).
2. E. O. Wilson, Sociobiology: The New Synthesis (Belknap Press
of Harvard Univ. Press, 2000).
3. R. M. Sapolsky, Science 308, 648–652 (2005).
4. S. J. Mooney, D. E. Peragine, G. A. Hathaway, M. M. Holmes,
Soc. Neurosci. 9, 108–117 (2014).
5. C. Sandi, J. Haller, Nat. Rev. Neurosci. 16, 290–304 (2015).
6. F. Wang, H. W. Kessels, H. Hu, Trends Neurosci. 37, 674–682
(2014).
7. E. Bonabeau, G. Theraulaz, J. L. Deneubourg, Bull. Math. Biol.
61, 727–757 (1999).
8. K. Kura, M. Broom, A. Kandler, Bull. Math. Biol. 78, 1259–1290
(2016).
9. L. A. Dugatkin, M. Druen, Proc. Biol. Sci. 271, S488–S489 (2004).
10. M. Y. Chou et al., Science 352, 87–90 (2016).
11. S. Trannoy, E. A. Kravitz, Fly 11, 129–138 (2017).
12. L. Mah, M. C. Arnold, J. Grafman, Am. J. Psychiatry 161,
1247–1255 (2004).
13. P. H. Rudebeck et al., Eur. J. Neurosci. 26, 2315–2326 (2007).
14. N. Fujii, S. Hihara, Y. Nagasaka, A. Iriki, Soc. Neurosci. 4, 73–84
(2009).
15. K. L. Hillman, D. K. Bilkey, Nat. Neurosci. 15, 1290–1297 (2012).
16. T. Hosokawa, M. Watanabe, J. Neurosci. 32, 7662–7671 (2012).
17. J. Sallet et al., Science 334, 697–700 (2011).
18. F. Wang et al., Science 334, 693–697 (2011).
19. G. Lindzey, H. Winston, M. Manosevitz, Nature 191, 474–476
(1961).
20. B. N. Armbruster, X. Li, M. H. Pausch, S. Herlitze, B. L. Roth,
Proc. Natl. Acad. Sci. U.S.A. 104, 5163–5168 (2007).
21. E. S. Boyden, F. Zhang, E. Bamberg, G. Nagel, K. Deisseroth,
Nat. Neurosci. 8, 1263–1268 (2005).
22. C. R. Yang, J. K. Seamans, J. Neurosci. 16, 1922–1935 (1996).
23. R. J. Blanchard, P. M. Wall, D. C. Blanchard, Horm. Behav. 44,
161–170 (2003).
24. W. B. Hoover, R. P. Vertes, Brain Struct. Funct. 212, 149–179
(2007).
25. T. B. Franklin et al., Nat. Neurosci. 20, 260–270 (2017).
26. J. Y. Lin, M. Z. Lin, P. Steinbach, R. Y. Tsien, Biophys. J. 96,
1803–1814 (2009).
27. S. Nabavi et al., Nature 511, 348–352 (2014).
28. V. Pascoli, M. Turiault, C. Lüscher, Nature 481, 71–75 (2011).
29. M. E. Walton, D. M. Bannerman, M. F. Rushworth, J. Neurosci.
22, 10996–11003 (2002).
30. A. Friedman et al., Cell 161, 1320–1333 (2015).
31. M. R. Bailey, E. H. Simpson, P. D. Balsam, Neurobiol. Learn.
Mem. 133, 233–256 (2016).
32. C. B. Holroyd, S. M. McClure, Psychol. Rev. 122, 54–83 (2015).
33. Y. Hsu, R. L. Earley, L. L. Wolf, Biol. Rev. Camb. Philos. Soc. 81,
33–74 (2006).
AC KNOWLED GME NTS
We thank Q. Li, H. Kessels, H. Li, and W. Li for critical comments on
the manuscript; D. Anderson for stimulating discussions that led to
the idea of the warm spot test; J. Zhu and K. Yuan for assistance
in experiments; K. Deisseroth and Z. Qiu for AAV-ChR2 constructs;
B. Roth for AAV-hM4D and AAV-hM3D constructs; C. Li and X. Gu
for advice on analysis of tetrode recording data; and X. Xu for
Matlab code for behavior annotation. This work was supported by
grants from the Ministry of Science and Technology of China
(2011CBA00400 and 2016YFA0501000), the National Natural
Science Foundation of China (91432108, 31225010, and 81527901),
and the Strategic Priority Research Program (B) of the Chinese
Academy of Sciences (XDB02030004) to H.H. All the data
necessary to understand and assess the conclusions of this
manuscript are available in the supplementary materials.
Computer codes are archived at www.dropbox.com/sh/
7pthozsrzj4vxr0/AACMaW9a8_5ITDumIJw_6Z_pa?dl=0. T.Z. and
H.Z. conducted most optogenetic and behavioral experiments
and designed the experiments with H.H. T.Z. performed in vivo
tetrode recording with the help of Y.C. and Z.Y. and conducted
optogenetic LTP and LTD experiments. Z.F. performed the warm
spot test and dominance transfer experiments. F.W. and H.L.
participated in tube test and viral injection experiments. L.Z.,
L.L., Y.Z., and Z.W. participated in analysis of in vivo tetrode
recordings. H.H. conceived the project and wrote the manuscript
with the help of T.Z. and H.Z.
SUPPLEMENTARY MATERIALS
www.sciencemag.org/content/357/6347/162/suppl/DC1
Materials and Methods
Figs. S1 to S12
Table S1
References (34–51)
Movies S1 to S5
18 February 2017; accepted 9 June 2017
10.1126/science.aak9726
7 of 7