Experience-dependent plasticity in an innate social behavior is mediated by hypothalamic LTP

Experience-dependent plasticity in an innate social

behavior is mediated by hypothalamic LTP

Stefanos Stagkourakis

a,1



, Giada Spigolon



, Grace Liu



, and David J. Anderson

a,c,1

Division of Biology and Biological Engineering 156-29, Tianqiao and Chrissy Chen Institute for Neuroscience, California Institute of Technology, Pasadena,

CA 91125;

Biological Imaging Facility, California Institute of Technology, Pasadena, CA 91125; and

HHMI, California Institute of Technology, Pasadena,

CA 91125

Contributed by David J. Anderson, August 25, 2020 (sent for review June 9, 2020; reviewed by Robert C. Malenka and Richard Mooney)

All animals can perform certain survival behaviors without prior

experience, suggesting a

“

hard wiring

”

of underlying neural circuits.

Experience, however, can alter the expression of innate behaviors.

Where in the brain and how such plasticity occurs remains largely

unknown. Previous studies have established the phenomenon of

“

aggression training,

”

in which the repeated experience of winning

successive aggressive encounters across multiple days leads to in-

creased aggressiveness. Here, we show that this procedure also

leads to long-term potentiation (LTP) at an excitatory synapse, de-

rived from the posteromedial part of the amygdalohippocampal

area (AHiPM), onto estrogen receptor 1-expressing (Esr1

)neurons

in the ventrolateral subdivision of the ventromedial hypothalamus

(VMHvl). We demonstrate further that the optogenetic induction of

such LTP in vivo facilitates, while optogenetic long-term depression

(LTD) diminishes, the behavioral effect of aggression training, im-

plying a causal role for potentiation at AHiPM

→

VMHvl

Esr1

synapses

in mediating the effect of this training. Interestingly,

∼

25% of inbred

C57BL/6 mice fail to respond to aggression training. We show that

these individual differences are correlated both with lower levels of

testosterone, relative to mice tha

t respond to such training, and with

a failure to exhibit LTP after aggression training. Administration of

exogenous testosterone to such nonaggressive mice restores both

behavioral and physiological plasticity. Together, these findings re-

veal that LTP at a hypothalamic circuit node mediates a form of

experience-dependent plasticity

in an innate social behavior, and a

potential hormone-dependent basi

s for individual differences in such

plasticity among genetically identical mice.

innate behaviors

|

long-term potentiation

|

ventromedial hypothalamus

|

testosterone

rains evolved to optimize the survival of animal species by

generating appropriate behavioral responses to both stable

and unpredictable features of the environment. Accordingly, two

major brain strategies for behavioral control have been selected.

In the first,

“

hard-wired

”

neural circuits generate rapid innate

responses to sensory stimuli that have remained relatively con-

stant and predictable over evolutionary timescales (1, 2). In the

second, neural circuits generate flexible responses to stimuli that

can change over an individual

’

s life span, through learning and

memory (3, 4).

One common view is that these two strategies are implemented

by distinct neuroanatomical structures and neurophysiological

mechanisms. According to this view, in the mammalian brain,

innate behaviors are mediated by evolutionarily ancient subcorti-

cal structures, such as the extended amygdala and hypothalamus,

which link specific sensory inputs to evolutionarily

“

prepared

”

motor outputs through relatively stable synaptic connections

(5). In contrast, learned behaviors are mediated by more recently

evolved structures, such as the cortex and hippocampus, which

compute flexible input

–

output mapping responses through syn-

aptic plasticity mechanisms (6). This view has been supported by

studies that have revealed distinct anatomical pathways through

which olfactory cues evoke learned vs. innate behaviors in both the

mouse (7

–

9) and in

Drosophila

(reviewed in ref. 10).

This view of distinct neural pathways for innate vs. learned

behaviors, however, is challenged by the case of behaviors that,

while apparently

“

instinctive,

”

can nevertheless be modified by

experience. For example, studies in rodents have shown that de-

fensive behaviors such as freezing can be elicited by both un-

conditional and conditional stimuli, the latter via Pavlovian

associative learning (reviewed in ref. 11). In this case, the pre-

vailing view argues for parallel pathways: Conditioned defensive

behavior is mediated by circuitry involving the hippocampus,

the thalamus, and the basolateral/central amygdala whereas

innate defensive responses to predators are mediated by the

medial amygdala (MeA)/bed nucleus of the stria terminalis

(BNST) and hypothalamic structures (reviewed in ref. 12).

Although the basolateral amygdala contains representations of

unconditioned aversive and appetitive stimuli, these representa-

tions are used as the cellular substrate for pairing with conditioned

stimuli (13, 14). Despite this segregation of learned and innate

defensive pathways, it remains possible that experience-dependent

influences on other innate behaviors may involve plasticity at

synapses that directly mediate instinctive behaviors.

We have investigated this issue usi

ng intermale offensive aggres-

sion in mice. While aggression has been considered by ethologists as

a prototypical innate behavior (15, 16), animals can be trained to be

more aggressive by repeated fighting experience (17

–

19). The

neural substrates and physiological mechanisms underlying this

form of experience-dependent plasticity remain unknown. In-

terestingly, inbred strains of laboratory mice exhibit individual

Significance

Modification of instinctive behaviors occurs through experience,

yet the mechanisms through whic

h this happens have remained

largely unknown. Recent studies have shown that potentiation

of aggression, an innate behavior, can occur through repeated

winning of aggressive encounters. Here, we show that syn-

aptic plasticity at a specific excitatory input to a hypothalamic

cell population is correlated with, and required for, the ex-

pression of increasingly higher levels of aggressive behavior

following aggressive experience. We additionally show that

the amplitude and persistence of long-term potentiation at this

synapse are influenced by serum testosterone, administration

of which can normalize individual differences in the expression

of intermale aggression among genetically identical mice.

Author contributions: S.S., G.S., and D.J.A. designed research; S.S. and G.S. performed

research; S.S., G.S., G.L., and D.J.A. analyzed data; and S.S., G.S., and D.J.A. wrote

the paper.

Reviewers: R.C.M., Stanford University School of Medicine; and R.M., Duke University.

The authors declare no competing interest.

This open access article is distributed under

Creative Commons Attribution-NonCommercial-

NoDerivatives License 4.0 (CC BY-NC-ND)

To whom correspondence may be addressed. Email: stefanos.stagkourakis@caltech.edu

or wuwei@caltech.edu.

This article contains supporting information online at

https://www.pnas.org/lookup/suppl/

doi:10.1073/pnas.2011782117/-/DCSupplemental

First published September 24, 2020.

www.pnas.org/cgi/doi/10.1073/pnas.2011782117

PNAS

October 13, 2020

vol. 117

no. 41

25789

–

25799

NEUROSCIENCE

Downloaded at California Institute of Technology on October 13, 2020

differences in this form of plasticity, with up to 25% of animals

failing to respond to aggression training (19). The biological basis

of this heterogeneity is not understood. Here, we provide data

supporting a plausible explanation for both experience-dependent

changes and individual differences in male aggressiveness, one

that links physiological plasticity at hypothalamic synapses to ag-

gressive behavior and sex hormone levels.

Results

Aggression Training Increases VMHvl

Esr1

Neuron Activity.

Aggression

levels escalate following repeated fighting experience (19, 20), an

effect termed here as

“

aggression training.

”

Using a five-

consecutive-day resident

–

intruder (5cdRI) assay (Fig. 1

aggression training was performed on a cohort of C57BL/6 Esr1-

Cre mice (

=

138). Most of these animals displayed increased

aggression levels that remained significantly elevated, relative to

pretraining animals, over a prolonged period of time (maximal

period tested, 3 mo) (Fig. 1

–

). This assay enabled comparison

of socially naive, vs. aggressive (AGG), and nonaggressive

(NON) mice, the latter of which represented

∼

23% of all males

tested (Fig. 1

). Interestingly, aggression levels in AGG mice

were found to plateau on the fourth and fifth day of the 5cdRI

assay (Fig. 1

–

To test whether aggression training involves plasticity in a

structure that mediates an innate form of aggression (21), we

initially focused on VMHvl

Esr1

(estrogen receptor 1-expressing

neurons in the ventrolateral subdivision of the ventromedial hy-

pothalamus) neurons, a group of cells highly overlapping with

VMHvl progesterone receptor-expressing neurons, both of which

are necessary and sufficient for eliciting attack in socially naive

animals (22

–

24). Using brain slice Ca

imaging, the average

baseline activity of VMHvl

Esr1

neurons was found to increase in

AGGs but not in NONs, following aggression training (Fig. 1

–

Voltage-clamp ex vivo VMHvl

Esr1

neuron recordings revealed a

significant increase in the frequency and amplitude of sponta-

neous excitatory postsynaptic currents (sEPSCs) in AGG mice

(Fig. 2

–

), relative to socially naive animals. In contrast,

voltage-clamp ex vivo VMHvl

Esr1

neuron recordings in slices from

NON mice revealed an increase in the frequency and amplitude of

spontaneous inhibitory postsynaptic currents (sIPSCs) (

SI Ap-

pendix

,Fig.S1

). These observations raised the possibility that

synaptic potentiation/depression mechanisms may be present in

VMHvl

Esr1

neurons (25).

We first investigated whether potentiation of an excitatory

input occurs on these cells. Anatomical studies previously iden-

tified a strong purely excitatory input to VMHvl

Esr1

cells, which

originates anatomically from the posteromedial part of the

amygdalohippocampal area (AHiPM) (also termed the posterior

amygdala [PA]) (26). In vivo optogenetic activation of AHiPM

evoked aggression (Fig. 2

–

), confirming recent reports (27, 28).

We found, moreover, that this effect is amplified in aggression-

experienced animals (Fig. 2

–

). Investigation of the functional

connectivity at AHiPM

→

VMHvl

Esr1

synapses in acute slices

using optogenetic activation of AHiPM inputs (Fig. 2

–

) in-

dicated that this projection is entirely excitatory, and at least

in part monosynaptic (

SI Appendix

, Fig. S2

), with an absence

of any evoked responses at the reversal potential for excitation

(holding voltage

=

0 mV, Fig. 2

Middle Row

) and reliable

photostimulation-evoked currents at the reversal potential for

inhibition (holding voltage

=

−

70 mV, Fig. 2

Bottom Row

These observations raised the question of whether potentiation

at AHiPM

→

VMHvl

Esr1

synapses underlies the observed increase

in the excitatory synaptic input onto VMHvl

Esr1

neurons fol-

lowing aggression training.

Plasticity at a Hypothalamic Synapse Following Aggression Training.

At synapses that can undergo long-term potentiation (LTP), the

postsynaptic response to stimulation of excitatory presynaptic

inputs largely depends on the ratio of

-methyl-

-aspartate

(NMDA) and

-amino-3-hydroxy-5-methyl-4-isoxazolepropionic

acid (AMPA) receptors (29). Therefore, we measured the AMPA/

NMDA ratio at AHiPM

→

VMHvl

Esr1

synapses before vs. after

aggression training (30, 31). This analysis revealed a significantly

higher AMPA/NMDA ratio in AGGs following such training,

compared to socially naive and NON mice (Fig. 2

). We also

investigated synaptic integration properties (32) in VMHvl

Esr1

neurons from socially naive, AGG (trained), and NON mice. We

observed depressing/static synaptic integration in socially naive

and NON mice, and facilitating synaptic integration in the

VMHvl

Esr1

neurons of AGG (trained) mice (Fig. 2

and

Changes in the AMPA/NMDA ratio and synaptic integration

properties are often accompanied by changes in neuronal mor-

phology and dendritic spine complexity, which can be indicative

of structural LTP (sLTP) (33). To investigate this possibility,

VMHvl

Esr1

neurons from which the AMPA/NMDA measure-

ments were acquired were filled with neurobiotin, and super-

resolution images for reconstruction were obtained using the

Airyscanning technique in a ZEISS LSM 880 (34). Analysis of

second-order dendritic segments identified a prominent increase

in dendritic complexity in VMHvl

Esr1

neurons from AGG (trained)

mice,incomparisontosociallynaiveandNONmice(Fig.3

–

These changes were reflected in most spine parameters mea-

sured, including density, branching points, volume, area, length,

and mean diameter (Fig. 3

–

). However, the principal feature

was an increase in the number of short-length spines, suggesting

they were newly generated during or after training. While it is not

yet clear whether these second-order dendrites receive input from

AHiPM, these observations provided morphological evidence of

VMHvl

Esr1

plasticity following aggression training. We therefore

further investigated plasticity using

more specific electrophysiological

protocols.

Experimental Induction of LTP and Long-Term Depression at AHiPM

→

VMHvl

Esr1

Synapses.

LTP and long-term depression (LTD) can be

experimentally induced in slices from brain areas typically asso-

ciated with higher cognitive processing, such as the hippocampus

(reviewed in ref. 35), but few studies have demonstrated that this

can occur in the hypothalamus (36), a site traditionally considered

the source of instincts.

To determine whether LTP can be experimentally induced at

AHiPM

→

VMHvl

Esr1

synapses ex vivo

we employed acute VMHvl

slices and used an optogenetics protocol composed of three

bouts of photostimulation of Chronos-expressing AHiPM termi-

nals, during which the postsynaptic VMHvl

Esr1

neuron (identified

by Cre-dependent expression of tdTomato) was voltage-clamped

at a depolarized membrane potential (

−

30 mV, Fig. 4

and

The choice of this Hebbian stimulation protocol was based on our

initial finding that combined pre- and postsynaptic depolarization

was necessary for induction

of potentiation at AHiPM

→

VMHvl

Esr1

synapses (

SI Appendix

,Fig.S3

). The 20 Hz stimulation fre-

quency of AHiPM terminals was chosen based on our pre-

vious demonstration that dire

ct optogenetic s

timulation of

VMHvl

Esr1

neurons at this frequency drives action potential

firing with 100% spike fidelity ex vivo as well as in vivo

without

depolarization block (22), and that it produces reliable synaptic

integration in VMHvl

Esr1

neurons (Fig. 2

and

). Whole-cell

voltage-clamp recordings

—

composed of 20-min baseline and

20-min follow-up

—

revealed that the majority of VMHvl

Esr1

neurons recorded in slices from socially naive, AGG, and NON

mice exhibited synaptic potentiation in response to this manipu-

lation (Fig. 4

). Comparison of the responses with the animals

’

aggression phenotypes revealed, however, that the dynamics of the

response, including its maximum amplitude and persistence, dif-

fered between groups, with synaptic potentiation in slices from

NON mice returning to baseline levels within the maximal period

tested (20 min, Fig. 4

). Based on the Hebbian conditions

25790

www.pnas.org/cgi/doi/10.1073/pnas.2011782117

Stagkourakis et al.

Downloaded at California Institute of Technology on October 13, 2020

required to evoke this form of synaptic potentiation, and the

similarity of its features to hippocampal LTP (37), we refer to this

form of plasticity as hypothalamic LTP.

We also investigated whether AHiPM

→

VMHvl

Esr1

synapses

can also express long-term synaptic depression (LTD), using a

longer stimulation protocol f

or activating AHiPM terminals

(10 min continuous stimulation at 1Hz, Fig. 4

). Similar to the

case of LTP, most VMHvl

Esr1

neurons expressed LTD of varying

amplitude and dynamics, in a manner that varied with the animals

’

aggression phenotypes (Fig. 4

and

). Interestingly, VMHvl

Esr1

cells from NON mice expressed higher amplitude LTD (Fig. 4

)

than cells from other groups.

These observations raised the question of whether LTP

and LTD can be induced at AHiPM

→

VMHvl

Esr1

synapses in

behaving animals by optogenet

ic stimulation or aggression

training. For optogenetic induction of LTP, we simulta-

neously depolarized both pre- and postsynaptic terminals

using different opsins with an overlap at 535 nm (38):

Chronos in AHiPM and ChrimsonR in VMHvl

Esr1

neurons

(Fig. 4

). Chronic implantation of a silicon probe optrode in

the VMHvl (

Materials and Methods

) (39) allowed the de-

tection of optically induced LTP or LTD as a change in

AHiPM stimulation-evoked loca

l field excitatory postsyn-

aptic potentials (fEPSPs, Fig. 4

–

). Application of the

Hebbian protocol in socially naive, freely moving mice led to the

robust expression of LTP in VMHvl (Fig. 4

) while the 10-min

long stimulation at 1 Hz led to robust expression of LTD (Fig.

). Although VMHvl

Esr1

neurons in silicon probe recordings

were identified by optogenetic photo-tagging of postsynaptic cells,

we cannot exclude that other classes of VMHvl neurons contribute

to the recorded fEPSPs.

To investigate whether LTP can be behaviorally induced

in vivo, fEPSPs were monitored during a 10-min baseline period

and then daily following aggression training, using optogenetic

Fig. 1.

Aggression learning alters baseline activity dynamics in VMHvl

Esr1

neurons. (

) Schematic of the experimental design using the five-consecutive-day

resident

–

intruder (5cdRI) test, with three follow-up dates, used for the study of the behavioral effect of aggression training. (

) Summary indicative of the

number of male animals exhibiting the two distinct aggression phenotypes (

=

138). (

) Quantification of the cumulative duration (in %) of aggression per trial

(

=

15 AGG mice per group, Kruskal

–

Wallis one-way ANOVA with uncorrected Dunn

’

s post hoc test,

<

0.0001 between day 1 and day 3 of the 5cdRI assay,

=

0.4819 between day 4 and day 5 of the 5cdRI assay,

=

0.0149 between day 1 and day 30). (

) Quantification of attack latency (in seconds) of aggression per trial

(

=

15 AGG mice per group, Kruskal

–

Wallis one-way ANOVA with uncorrected Dunn

’

s post hoc test,

<

0.0001 between day 1 and day 3 of the 5cdRI assay,

=

0.5602 between day 4 and day 5 of the 5cdRI assay,

=

0.2184 between day 1 and day 30). (

) Quantification of the average attack episode duration (in seconds)

per trial (

=

15 AGG mice per group, Kruskal

–

Wallis one-way ANOVA with uncorrected Dunn

’

s post hoc test,

<

0.0001 between day 1 and day 3 of the 5cdRI

assay,

=

0.3326 between day 4 and day 5 of the 5cdRI assay,

=

0.0209 between day 1 and day 30). (

) Behavior raster plots from AGG and NON mice, at

different days of the 5cdRI test. (

) Baseline Ca

activity of VMHvl

Esr1

neurons recorded ex vivo, in brain slices of socially naive, aggressive (AGG), and nonag-

gressive (NON) males. (

) Quantification of active cells per slice (

=

16 to 19 brain slices, collected from

=

7 to 9 mice, one-way ANOVA with Dunnett

’

sposthoc

test,

=

0.0002 between socially naive and AGG mouse brain slices,

=

0.6358 between socially naive and NON mouse brain slices). (

) Quantification of Ca

spike

frequency per cell (

=

16 to 19 brain slices, collected from

=

7 to 9 mice, Kruskal

–

Wallis one-way ANOVA with Dunn

’

s post hoc test,

<

0.0001 between socially

naive and AGG mice,

=

0.3331 between socially naive and NON mice). (

) Quantification of Ca

spike amplitude (

=

16 to 19 brain slices, collected from

=

7to

9 mice, Kruskal

–

Wallis one-way ANOVA with Dunn

’

sposthoctest,

<

0.0001 between socially naive and AGG mice,

<

0.0001 between socially naive and NON

mice). ns, not significant; *

<

0.05, ***

<

0.001, ****

<

0.0001.

F/F was used to calculate Ca

-dependent changes in fluorescence, where F is the instan-

taneous fluorescence from the raw ROI time series. In box plots, the median is represented by the center line, the interquartile range is represented b

y the box

edges, the bottom whisker extends to the minimal value, and the top whisker extends to the maximal value.

Stagkourakis et al.

PNAS

October 13, 2020

vol. 117

no. 41

25791

NEUROSCIENCE

Downloaded at California Institute of Technology on October 13, 2020

test pulses to briefly activate AHiPM terminals. Indeed, LTP

(measured as an increase in fEPSP amplitude) was induced in

VMHvl

Esr1

neurons immediately after aggression (Fig. 4

–

=

4 mice tested). Notably, the behavioral induction of LTP

(Fig. 4

) led to a persistent change in the amplitude of the

fEPSP. This might suggest a lack of an early- vs. late-phase

Fig. 2.

AHiPM

→

VMHvl

Esr1

synapses become potentiated following aggression training. (

) Representative recordings of spontaneous excitatory postsynaptic

currents (sEPSCs) from VMHvl

Esr1

neurons, from socially naive, AGG, and NON mice. (

Left

) Cumulative frequency distribution plot of sEPSC IEI in voltage-

clamp recordings collected from VMHvl

Esr1

neurons from socially naive, AGG, and NON mice (

=

14 to 18 VMHvl

Esr1

neuron recordings per group, collected from 8

to 10 mice per group, Kolmogorov

–

Smirnov test,

<

0.0001 between socially naive and AGG mice,

=

0.3454 between socially naive and NON mice). (

Right

)

Comparison of sEPSC frequency from voltage-clamp recordings collected from VMHvl

Esr1

neurons from socially naive, AGG, and NON mice (

=

14 to 18 VMHvl

Esr1

neuron recordings per group, collected from 8 to 10 mice per group, one-way ANOVA with Dunnett

’

sposthoctest,

<

0.0001 between socially naive and AGG

mouse brain slices,

=

0.2576 between socially naive and NON mouse brain slices). (

Left

) Cumulative frequency distribution plot of sEPSC amplitude in voltage-

clamp recordings collected from VMHvl

Esr1

neurons from socially naive, AGG, and NON mice (

=

14 to 18 VMHvl

Esr1

neuron recordings per group, collected from 8

to 10 mice per group, Kolmogorov

–

Smirnov test,

<

0.0001 between socially naive and AGG mice,

<

0.0001 between socially naive and NON mice). (

Right

)

Comparison of sEPSC frequency from voltage-clamp recordings collected from VMHvl

Esr1

neurons from socially naive, AGG, and NON mice (

=

14 to 18 VMHvl

Esr1

neuron recordings per group, collected from 8 to 10 mice per group, Kruskal

–

Wallis one-way ANOVA with uncorrected Dunn

’

s post hoc test,

=

0.0041 between

socially naive and AGG mouse brain slices,

=

0.6712 between socially naive and NON mouse brain slices). (

Left

) Schematic of the experimental design used for

optogenetic studies of aggression following photoactivation of AHiPM. (

Right

) Confocal image indicative of Chronos-eYFP expression in AHiPM. (

)Schematic

illustration of the experimental protocol used in AHiPM

Chronos

stimulation experiments. (

) Sample behavior raster plots with in vivo optogenetics and social

behavior in the resident

–

intruder (RI) assay, of socially naive and AGG mice. (

) Quantification of attack latency, in the first and sixth RI trial (

=

8 mice per group,

first RI, two-sided Mann

–

Whitney

test,

=

0.0033 between YFP and Chronos groups, sixth RI, two-tailed unpaired

test,

=

0.0022 between YFP and Chronos

groups). (

) Quantification of attack duration, in the first and sixth RI trial (

=

8 mice per group, first RI, two-sided Mann

–

Whitney

test,

=

0.0079 between YFP

and Chronos groups, sixth RI,

=

0.0011 between YFP and Chronos groups). (

Top

) Schematic of the experimental design used for the study of the

AHiPM

→

VMHvl synapse. (

Bottom

) Confocal image indicative of AHiPM originating processes in VMHvl. (

) Identification of the AHiPM

→

VMHvl synapse as purely

excitatory, and extraction of the AMPA to NMDA ratio in socially naive, AGG, and NON mice (average of

=

13 to 14 neuron recordings from 8 to 9 socially naive,

AGG, and NON mice, respectively). (

) Quantification of the AMPA/NMDA ratio (

=

13 to 14, Kruskal

–

Wallis one-way ANOVA with Dunn

’

s post hoc test,

=

0.0005 between socially naive and AGG mice,

>

0.9999 between socially naive and NON mice). (

) Synaptic integration in VMHvl

Esr1

neurons from socially naive,

AGG, and NON mice (average traces of

=

7 to 10 neuron recordings from 7 to 9 mice, respectively). (

) Quantification of the five optically evoked excitatory

postsynaptic potentials (oEPSPs) peak amplitude presented in

. (Top) oEPSP amplitude quantification in VMHvl

Esr1

neurons recorded from socially naive mice (

=

10 neurons from nine mice, Friedman one-way ANOVA with Dunn

’

s post hoc test,

>

0.9999 between first and second pulse,

=

0.3587 between first and third

pulse,

=

0.0028 between first and fourth pulse, and

=

0.0009 between first and fifth pulse). (

Middle

) oEPSP amplitude quantification in VMHvl

Esr1

neurons

recorded from AGG mice (

=

10 neurons from nine mice, one-way ANOVA with Dunnett

’

s post hoc test,

=

0.2935 between first and second pulse,

<

0.0001

between first and third pulse,

<

0.0001 between first and fourth pulse, and

<

0.0001 between first and fifth pulse). (

Bottom

) oEPSP amplitude quantification in

VMHvl

Esr1

neurons recorded from NON mice (

=

7 neurons from seven mice, one-way ANOVA with Dunnett

’

sposthoctest,

=

0.9865 between first and second

pulse,

=

0.5704 between first and third pulse,

=

0.0751 between first and fourth pulse, and

=

0.9803 between first and fifth pulse). ns, not significant; **

<

0.01, ***

<

0.001, ****

<

0.0001. In box plots, the median is represented by the center line, the interquartile range is represented by the box edges, the bottom

whisker extends to the minimal value, and the top whisker extends to the maximal value. V

, holding voltage; IEI, inter-event interval, o.f., optic fiber; DIO-tdT,

Cre-dependent tdTomato expression.

25792

www.pnas.org/cgi/doi/10.1073/pnas.2011782117

Stagkourakis et al.

Downloaded at California Institute of Technology on October 13, 2020

distinction in LTP at AHiPM

→

VMHvl

Esr1

synapses, in contrast

to LTP in the hippocampus and the amygdala (40).

The above findings identified hy

pothalamic synaptic plastic-

ity, and specifically LTP and LTD

, as mechanisms that can alter

VMHvl

Esr1

neuronal excitability and w

hich are correlated with

behavioral plasticity. Next, we

sought to address whether LTP

or LTD play a causal role in the behavioral effect of aggression

training.

LTP Facilitates and LTD Inhibits Potentiation of Aggression Following

Training.

To address whether LTP at AHiPM

→

VMHvl

Esr1

syn-

apses can influence the expression of aggressive behavior in in-

experienced animals, we optogenetically induced LTP in vivo in

socially naive, solitary mice. The AHiPM of Esr1-Cre mice was

transduced with Chronos or YFP while VMHvl

Esr1

neurons were

transduced with Cre-dependent ChrimsonR or mCherry (Fig. 4

and

SI Appendix

, Fig. S4

). The effects of these manipulations

Fig. 3.

Increased dendritic spine complexity in VMHvl

Esr1

neurons following aggression training. (

) Maximum projection confocal image of a VMHvl

Esr1

neuron from a socially naive mouse recorded ex vivo

and filled with Neurobiotin. (

) Three-dimensional (3D) rendering of a s

econd order dendritic segment Airyscan

image from the neuron presented in

) Overlay of reconstruction data generated in Imaris aga

inst 3D rendering for the dendritic segment presented in

)

Reconstructed dendritic segment of a VMHvl

Esr1

neuron from a socially naive mouse, with color coding for the dendrite and spines. (

) Maximum projection confocal

image of a VMHvl

Esr1

neuron from an aggressive (AGG) mouse recorded ex vivo

and filled with Neurobiotin. (

) A 3D rendering of a second order dendritic segment

Airyscan image from the neuron presented in

) Overlay of reconstruction data generated in Imaris against 3D rendering for the dendritic segment presented in

(

) Reconstructed dendritic segment of a VMHvl

Esr1

neuron from an AGG mouse, with color coding for the dendrite and spines. (

) Maximum projection confocal image

of a VMHvl

Esr1

neuron from a nonaggressive (NON) mouse recorded ex vivo

and filled with Neurobiotin. (

) A 3D rendering of a second order dendritic segment

Airyscan image from the neuron presented in

) Overlay of reconstruction data generated in Imaris against 3D rendering for the dendritic segment presented in

(

) Reconstructed dendritic segment of a VMHvl

Esr1

neuron from a NON mouse, with color coding for the dendrite and spines. (

) Quantification of spine density in

second order dendrites of VMHvl

Esr1

neurons from socially naive, AGG, and NON mice (

=

3 to 5 cells per group,

=

1 cell per brain slice per animal,

=

23 to 26

segments analyzed per group, Kruskal

–

Wallis one-way ANOVA with Dunn

’

s post hoc test,

<

0.0001 between socially naive and AGG mice,

=

0.0432 between

socially naive and NON mice). (

) Quantification of branching points in second order dendrites of VMHvl

Esr1

neurons from socially naive, AGG, and NON mice

(

=

23 to 26 segments analyzed per group, Kruskal

–

Wallis one-way ANOVA with Dunn

’

s post hoc test,

<

0.0001 between socially naive and AGG mice,

=

0.9969 between socially naive and NON mice). (

) Quantification of spine volume in second order dendrites of VMHvl

Esr1

neurons from socially naive, AGG, and

NON mice (

=

23 to 26 segments analyzed per group, Kruskal

–

Wallis one-way ANOVA with Dunn

’

s post hoc test,

<

0.0001 between socially naive and AGG

mice,

=

0.4640 between socially naive and NON mice). (

) Frequency distribution plot of spine area, of spines present in second order dendrites in VMHvl

Esr1

neurons from socially naive, AGG, and NON mice (

=

3 to 5 cells per group,

=

1 cell per brain slice per animal,

=

402 to 2,365 spines per group). (

) Frequency

distribution plot of spine length, of spines present in second order dendrites in VMHvl

Esr1

neurons from socially naive, AGG, and NON mice (

=

402 to 2,365

spines per group). (

) Frequency distribution plot of spine mean diameter, of spines present in second order dendrites in VMHvl

Esr1

neurons from socially naive,

AGG, and NON mice (

=

402 to 2,365 spines per group). ns, not significant; *

<

0.05, ****

<

0.0001. In box plots, the median is represented by the center line,

the interquartile range is represented by the box edges, the bottom whisker extends to the minimal value, and the top whisker extends to the maximal value.

Stagkourakis et al.

PNAS

October 13, 2020

vol. 117

no. 41

25793

NEUROSCIENCE

Downloaded at California Institute of Technology on October 13, 2020

Fig. 4.

Induction of LTP and LTD at AHiPM

→

VMHvl

Esr1

synapses ex vivo and in vivo. (

) Schematic of the experimental design used to study the induction of

LTP and LTD ex vivo in socially naive, aggressive (AGG) and nonaggressive (NON) mice. (

) Illustration of the experimental protocol used to induce LTP in the

AHiPM

→

VMHvl synapse. (

Left

) Heat map illustrating the magnitude of LTP induction in all recorded VMHvl

Esr1

neurons. (

Middle

) Average current im-

mediately prior to and following the induction of LTP (

Middle

=

28 neurons collected from the three groups

—

socially naive, AGG, and NON, with

=

6to8

mice per group, light color envelope is the SE). (

Right

) Quantification of the optically evoked excitatory post-synaptic current (oEPSC), prior to and following

the induction of LTP (pre- vs. postpairing,

=

28 neurons collected from the three groups

—

socially naive, AGG, and NON, with

=

6 to 8 mice per group, two-

tailed Wilcoxon signed-rank test,

<

0.0001). (

) Identification of differences in amplitude and persistence of LTP in socially naive, AGG, and NON mice (

=

neurons for six socially naive mice,

=

10 neurons from eight AGGs, and

=

9 neurons from six NONs, Kolmogorov

–

Smirnov test for curve comparison,

=

0.0183 between socially naive and AGG mice,

<

0.0001 between socially naive and NON mice, and

<

0.0001 between AGG and NON mice). (

) Illustration

of the experimental protocol used to induce LTD in the AHiPM

→

VMHvl synapse. (

Left

) Heat map illustrating the magnitude of LTD induction in all recorded

VMHvl

Esr1

neurons. (

Middle

) Average current immediately prior to and following the induction of LTD (

Middle

=

33 neurons collected from the three

groups

—

socially naive, AGG, and NON, with

=

8 mice per group, light color envelope is the SE). (

Right

) Quantification of the oEPSC, prior to and following

the induction of LTD (pre- vs. postpairing,

=

33 neurons collected from the three groups

—

socially naive, AGG, and NON, with

=

8 mice per group, two-

tailed Wilcoxon signed-rank test,

<

0.0001). (

) LTD dynamics in the three groups (

=

12 neurons for eight socially naive mice,

=

10 neurons from eight

AGGs, and

=

11 neurons from eight NONs, Kolmogorov

–

Smirnov test for curve comparison,

=

0.0002 between socially naive and AGG mice,

>

0.9999

between socially naive and NON mice, and

=

0.0008 between AGG and NON mice). (

) Schematic of the experimental design used to study the induction

of LTP and LTD in vivo in socially naive mice. (

) Schematic illustration of the target coordinates of the optrode used to record local field potentials in VMHvl.

(

Left

) Representative confocal image of Chronos-eYFP expression in AHiPM. (

Middle

) Representative confocal image of the silicon probe tract targeted to

VMHvl. (

Right

) High magnification confocal image of VMHvl. (

) Illustration of the experimental design used to induce LTP or LTD in the AHiPM

→

VMHvl

synapse in vivo. (

Left

) Plot of the average of four experiments from four mice of field EPSP slope (normalized to baseline period) before and after optically

induced LTP (oiLTP). (Middle) In vivo average field response prior to and following the induction of LTP. (

Right

) Quantification of optically induced fEPSPs,

prior to and following the induction of LTP (pre- vs. postpairing,

=

4 mice per group, two-tailed paired

test,

=

0.0283). (

Left

) Plot of the average of

four experiments from four mice of fEPSP slope (normalized to baseline period) before and after oiLTD. (

Middle

) In vivo average field response prior to and

following the induction of LTD. (

Right

) Quantification of optically induced fEPSPs, prior to and following the induction of LTD (pre- vs. postpairing,

=

mice

per group, two-tailed paired

test,

=

0.0007). (

) Illustration of the experimental design used to test the behavioral induction of LTP. (

Left

)Plotofthe

average of four experiments from four mice of fEPSP slope (normalized to baseline period) before and after behaviorally induced LTP (biLTP). (

Middle

) In vivo

average field response prior to and following the behavioral induction of LTP. (

Right

) Quantification of optically induced fEPSPs, prior to and following social

behavior experience in a socially naive mouse (pre- vs. postpairing,

=

4 mice per group, two-tailed paired

test,

=

0.0071). (

) Illustration of the behaviors

expressed in the resident

–

intruder assay from socially naive mice used for the in vivo study of hypothalamic LTP. ns, not significant; *

<

0.05, **

<

0.01,

***

<

0.001, ****

<

0.0001. In box plots, the median is represented by the center line, the interquartile range is represented by the box edges, the bottom

whisker extends to the minimal value, and the top whisker extends to the maximal value.

25794

www.pnas.org/cgi/doi/10.1073/pnas.2011782117

Stagkourakis et al.

Downloaded at California Institute of Technology on October 13, 2020

were investigated behaviorally, not physiologically; therefore,

silicon probes were not implanted.

Following the application of the optogenetic LTP induction

protocol over three consecutive days in the absence of any social

experience, the RI test was performed on the fourth day without

optogenetic manipulations (

SI Appendix

, Fig. S4

). The opsin-

expressing mice exhibited significantly elevated levels of ag-

gression, measured as an increase in attack duration, number of

attacks per trial, and a decrease in attack latency, in comparison

to YFP-expressing control mice (

SI Appendix

, Fig. S4

–

These data indicate that the experimental induction of LTP at

AHiPM

→

VMHvl

Esr1

synapses in vivo can increase aggressive-

ness in the absence of social experience.

We investigated next whether optogenetically evoked LTP or

LTD can modify the effects of aggression training (Figs. 4

–

and 5

). To determine whether LTP could facilitate aggression

training, the LTP induction protocol was delivered at the end of

each RI trial, in both the control (XFP-expressing) and LTP

(opsin-expressing) groups. In a separate experiment, the LTD

induction protocol was delivered at the end of each RI trial in

both the control and opsin-expressing groups (Fig. 5

). Appli-

cation of LTD is expected to override any endogenous LTP that

may have occurred (41); therefore, if LTP is required in vivo for

the behavioral effect of aggression training, then LTD induction

should diminish this effect. As in Fig. 1

, smaller size BALB/c

intruders were used as intruders in the RI assay to increase the

expression of aggression by C57BL/6N

Esr1-Cre

resident mice (Fig. 5

Aggression levels were recorded and analyzed on each day of

the 5cdRI assay (i.e., 24 h following the previous LTP or LTD

manipulation), with the exception of day 1.

Applying LTP or LTD induction protocols in vivo facilitated

or diminished the behavioral effect of aggression training, re-

spectively, as reflected by increased aggression/trial (percentage

of the total trial duration occupied by aggressive behavior) and

decreased attack latency (Fig. 5

–

). Interestingly, although

LTP was found to enhance the behavioral effect of aggression

training on the second and third day of the 5cdRI assay, the

effect plateaued on days 4 and 5, at a level similar to the control

group (see also Fig. 1

), suggesting an effect that can increase

the rate but not the extent of learning. In contrast, LTD had a

profound inhibitory effect on aggression training, leading to similar

aggression levels following day 1 and day 5 of the 5cdRI test (Fig. 5

two-tailed paired

test,

=

0.0592 between day 1 and day 5 in the

LTD group).

We investigated next whether the plateau in the effect of

optogenetically evoked LTP following training day 3 is indeed

due to a

“

ceiling effect

”

in the aggression training paradigm. To

do this, we performed a modified RI test following completion of

the 5cdRI training routine. On day 6, control or LTP-induced

residents were tested in a novel arena against a larger CD1 male

intruder (Fig. 5

and

). Under these conditions, aggressive

resident mice are less likely to attack the intruder (42). We

reasoned that, if LTP mice reached

“

ceiling

”

levels of aggression

in the standard 5cdRI assay, using subordinate BALB/c in-

truders, they might nevertheless show higher aggression toward

the CD1 intruders.

Indeed, under these conditions, the 5cdRI/LTP-treated group

exhibited significantly higher aggression levels toward CD1 in-

truders than any other tested group while the control and LTD

groups expressed similarly low aggression levels (Fig. 5

–

This finding suggests that experimental induction of hypotha-

lamic LTP in VMHvl

Esr1

neurons can facilitate attack under

modified RI conditions where resident aggressiveness is other-

wise reduced, relative to conventional RI assays.

Together these experiments demonstrate a potential role for

LTP and LTD in aggression plasticity. We next investigated the

basis for individual differences in aggression training among

genetically identical mice, by asking whether we could identify

any experimental intervention that would allow aggression and/

or hypothalamic LTP to be expressed in NON mice.

Testosterone Enables the Expression of Aggression and Hypothalamic

LTP in NON Mice.

Levels of testosterone (T) correlate with ag-

gression and dominance in numerous species (43, 44) while the

administration of T in female mice, or following castration in

male mice, introduces aggression to the animal

’

s behavioral

repertoire (45, 46). As T levels are subject to environmental

influences (47), we sought to determine whether individual dif-

ferences in levels of the hormone were detectable among ge-

netically identical, inbred C57BL/6N mice and, if so, whether

they correlated with, and were responsible for, individual dif-

ferences in the capacity to undergo aggression training.

To investigate whether serum T levels differ between NON

and AGG mice, we collected blood samples at different time

points during the 5cdRI test (Fig. 6

–

). This experiment

revealed retrospectively that, prior to fighting, a small but sta-

tistically significant (

<

0.05) difference in serum T was de-

tectable between NON and AGG mice (Fig. 6

). This difference

was further accentuated following aggression training (Fig. 6

Serum T levels remained unaltered in NON mice while they

increased in AGG mice (Fig. 6

). Interestingly, the increase in

serum T in AGG mice occurred in the first 3 d and was not further

accentuated through additional aggression training (Fig. 6

Together, these data reveal a correlation between individual dif-

ferences in T and the ability to respond to aggression training in

NON vs. AGG mice, as well as between levels of aggressiveness

and T levels in AGG mice during training.

To test whether T levels are causally responsible for the dif-

ference in aggressiveness between NON and AGG mice, sub-

cutaneous (s.c.) osmotic minipumps containing T or vehicle were

implanted in NON mice identified by their failure to show ag-

gressiveness following 5cdRI training (Fig. 6

). The serum T

levels in NONs measured at 7 d post-T minipump implantation

were significantly higher than (but within the upper quartile of)

the maximum T levels measured in AGG mice (Fig. 6

and

)

(serum T in NONs following T administration through mini-

pump 39.22

±

2.73 ng/mL, serum T in AGGs following aggres-

sion training 16.61

±

1.01 ng/mL,

=

6 and 46 mice, respectively,

two-sided Mann

–

Whitney

test,

<

0.0001). Strikingly, the

administration of exogenous T induced aggression in all NON

mice tested (Fig. 6

–

). We investigated next whether T ad-

ministration influenced the induction and/or expression of LTP,

using ex vivo recordings in acute VMHvl slices from vehicle

vs. T-treated NON mice. Using the Hebbian induction pro-

tocol, stronger LTP could be elicited in VMHvl

Esr1

neurons

recorded in slices from T-treated than from control NON mice

(Fig. 6

–

). Thus, T implants facilitate LTP induction ex vivo in

NON mice.

An important remaining question was whether LTP was

expressed in VMHvl

Esr1

neurons in vivo, following aggression

training in T-treated NON mice. To address this question, we

used the design previously described in Fig. 4

and

, in which

AHiPM was transduced with Chronos, while VMHvl

Esr1

cells

were transduced with ChrimsonR. A novel BALB/c, small size

male intruder was introduced into the NON

’

s home cage

(Fig. 6

). In vehicle-treated mice, social interactions with intruder

mice, but no aggression, were observed, and LTP did not occur

in vivo, as measured by fEPSP recordings in response to opto-

genetic stimulation of Chronos-expressing AHiPM terminals

(Fig. 6

–

). However, T administration through s.c. osmotic

minipumps led to the expression of both aggressive behavior and

in vivo behaviorally induced LTP, in NON mice (Fig. 6

–

These findings suggest that individual differences in serum T

are responsible, at least in part, for individual differences in the

capacity for aggression training among inbred mice. Elevation of

serum T in NON mice confers susceptibility to aggression training, as

Stagkourakis et al.

PNAS

October 13, 2020

vol. 117

no. 41

25795

NEUROSCIENCE

Downloaded at California Institute of Technology on October 13, 2020

Fig. 5.

Optogenetic induction of LTP or LTD at AHiPM

→

VMHvl

Esr1

synapses in vivo facilitates or abolishes, respectively, the effect of aggression training. (

Left

) Representative confocal image and schematic indicative of ChrimsonR expression in VMHvl

Esr1

neurons, eYFP terminals of the AHiPM

→

VMHvl projection,

and the optic fiber tract terminating above VMHvl. (

Right

) Representative confocal image and schematic indicative of Chronos-eYFP expression in AHiPM. (

)

Schematic of the experimental design used to identify whether LTP and LTD have an impact on aggression training. (

) Weight measurements of the mice

which were used in the protocol; specifically, the Esr1-Cre mice were used as residents, the BALB/c as intruders, and the CD1 as novel conspecifics in a novel/

neutral arena (

=

32 to 64 mice per group, one-way ANOVA with Tukey

’

s test,

<

0.0001 between Esr1-Cre and BALB/c mice,

<

0.0001 between Esr1-Cre

and CD1 mice). (

) Quantification of aggression levels expressed during a trial throughout the 5cdRI test in the YFP (control), LTP, and LTD groups. (

)

Quantification of aggression levels on the first day of the 5cdRI test (

=

8 mice per group, two-tailed unpaired

test,

=

0.1049 between YFP and LTP groups,

=

0.2304 between YFP and LTD groups). (

) Quantification of aggression levels on the third day of the 5cdRI test (

=

8 mice per group, two-tailed unpaired

test,

=

0.0162 [observed power

=

0.989, Cohen

’

=

0.7979, difference between means

=

9.13

±

3.34%, 95% CI

=

1.966 to 16.29] between YFP [lower 95%

=

6.452, higher 95% CI

=

16.06] and LTP [lower 95% CI

=

14.12, higher 95% CI

=

26.65] groups,

=

0.0017 between YFP and LTD groups). (

) Quantification

of aggression levels on the fifth day of the 5cdRI test (

=

8 mice per group, two-tailed unpaired

test,

=

0.0777 between YFP and LTP groups,

<

0.0001

between YFP and LTD groups). (

) Quantification of attack latency throughout the 5cdRI test in the YFP (control), LTP, and LTD groups. (

) Quantification of

attack latency on the first day of the 5cdRI test (

=

8 mice per group, two-tailed unpaired

test,

=

0.1406

between YFP and LTP groups,

=

0.3688 between

YFP and LTD groups). (

) Quantification of attack latency on the third day of the 5cdRI test (

=

8 mice per group, two-sided Mann

–

Whitney

test,

=

0.0415

[observed power

=

0.999, Cohen

’

=

0.6072, difference between means

=

159.40

±

90.79 s, 95% CI

=

−

378.8 to 60.04] between YFP [lower 95% CI

=

−

21.05,

higher 95% CI

=

407.0] and LTP [lower 95% CI

=

16.54, higher 95% CI

=

50.56] groups,

=

0.0019 between YFP and LTD groups). (

) Quantification of attack

latency on the fifth day of the 5cdRI test (

=

8 mice per group, two-sided Mann

–

Whitney

test,

=

0.5054 between YFP and LTP groups, two-tailed unpaired

test,

=

0.0052 between YFP and LTD groups). (

) Representative behavior raster plots of YFP, LTP, and LTD mouse behavior in a novel arena toward a novel

CD1 conspecific. (

) Quantification of aggression levels on the sixth day against a CD1 male (

=

8 mice per group, two-tailed unpaired

test,

=

0.0387

[observed power

=

0.999, Cohen

’

=

0.8980, difference between means

=

9.816

±

3.864%, 95% CI

=

0.6784 to 18.95] between YFP [lower 95% CI

=

0.8293,

higher 95% CI

=

5.768] and LTP [lower 95% CI

=

5.190, higher 95% CI

=

21.04] groups, two-sided Mann

–

Whitney

test,

=

0.0295

[observed power

=

0.907,

Cohen

’

=

0.7357, difference between means

=

2.161

±

1.069%, 95% CI

=

−

4.616 to 0.2938] between YFP [lower 95% CI

=

0.1860, higher 95% CI

=

5.052]

and LTD groups [lower 95% CI

=

−

0.2284, higher 95% CI

=

1.144]). (

) Quantification of attack latency on the sixth day against a CD1 male (

=

8 mice per

group, two-tailed unpaired

test,

=

0.0328 [observed power

=

0.985, Cohen

’

=

1.0431, difference between means

=

227.00

±

82.23 s, 95% CI

=

25.81 to

428.2] between YFP [lower 95% CI

=

67.01, higher 95% CI

=

477.9] and LTP groups [lower 95% CI

=

4.196, higher 95% CI

=

86.63], two-sided Mann

–

Whitney

test,

>

0.9999 between YFP and LTD groups). (

) Quantification of close investigation on the sixth day against a CD1 male (

=

8 mice per group, two-tailed

unpaired

test,

=

0.6973 between YFP and LTP groups,

=

0.6158 between YFP and LTD groups). ns, not significant; *

<

0.05, **

<

0.01, ****

<

0.0001. In

box plots, the median is represented by the center line, the interquartile range is represented by the box edges, the bottom whisker extends to the minimal

value, and the top whisker extends to the maximal value.

25796

www.pnas.org/cgi/doi/10.1073/pnas.2011782117

Stagkourakis et al.

Downloaded at California Institute of Technology on October 13, 2020

Fig. 6.

Testosterone administration leads to the expression of hypothalamic LTP and aggression in previously nonaggressive males. (

) Schematic of the

experimental design used to identify aggressive (AGG) and nonaggressive (NON) males, from which tails blood samples were collected for quantification of

serum testosterone levels. (

) Serum testosterone levels in NON vs. AGG mice prior to any aggression experience (

=

24 to 36 samples per group, two-sided

Mann

–

Whitney

test,

=

0.0203 between NON and AGG groups). Mice were assigned as NON or AGG, according to whether they expressed aggression on

the first day of the 5cdRI test. (

) Serum testosterone levels in NON vs. AGG mice after completion of the 5cdRI test (

=

14 to 46 samples per group, two-tailed

unpaired

test,

<

0.0001 between NON and AGG groups). Mice that did not express any aggression/attack behavior throughout the 5cdRI test were assigned

to the NON group. All other mice were included in the AGG group. (

Left

) Quantification of serum testosterone levels in NON mice throughout the 5cdRI

test (

=

14 to 24 samples per group, Kruskal

–

Wallis one-way ANOVA with Dunn

’

s post hoc test,

>

0.9999 between

“

prior to 5cdRI

”

and

“

following day 3,

”

>

0.9999 between

“

prior to 5cdRI

”

and

“

following day 5,

”

and

>

0.9999 between

“

following day 3

”

and

“

following day 5

”

groups). (

Right

) Quantification

of serum testosterone levels in AGG mice throughout the 5cdRI test (

=

36 to 46 samples per group, one-way ANOVA with Tukey

’

s test,

<

0.0001 between

“

prior to 5cdRI

”

and

“

following day 3,

”

<

0.0001 between

“

prior to 5cdRI

”

and

“

ing day 5,

”

and

=

0.7060 between

“

following day 3

”

and

“

following day 5

”

groups). (

) Schematic of the experimental design used to identify NON mice and perform s.c. testosterone minipump implantation. (

)

Serum testosterone levels in control vs. testosterone-treated mice (

=

6 mice per group, two-tailed unpaired

test,

<

0.0001 between vehicle and tes-

tosterone). (

) Representative behavior raster plots of vehicle vs. testosterone-treated mice. (

) Quantification of the number of mice that switched

aggression phenotype following vehicle vs. testosterone administration (

=

0/6 in the vehicle-treated group vs.

=

6/6 in the testosterone-treated group,

two-sided Mann

–

Whitney

test,

=

0.0022 between vehicle and testosterone). (

) Quantification of attack duration (

=

6 mice per group, two-sided

Mann

–

Whitney

test,

=

0.0022 between vehicle and testosterone). (

) Quantification of attack frequency (number (#) of attacks per trial;

=

6 mice per

group, two-sided Mann

–

Whitney

test,

=

0.0022 between vehicle and testosterone). (

) Schematic of the experimental design used to study the induction

and modulation of LTP by testosterone in the AHiPM

→

VMHvl synapse in brain slices from NON mice. Note that slices were taken from animals that received

T injections, but no behavioral training or other social experience. (

) Schematic of the LTP induction protocol, utilizing simultaneous photostimulation of the

AHiPM terminals in VMHvl through the opsin Chronos and depolarization of the VMHvl

Esr1

neuron through voltage clamp at

−

30 mV. (

) Heat map il-

lustrating the magnitude of LTP induction in VMHvl

Esr1

neurons from vehicle- vs. testosterone-treated mice. (

) Average current immediately prior to and

following the induction of LTP in vehicle vs. testosterone conditions (light color envelope is the SE). (

) Quantification of the oEPSC, prior to and following the

induction of LTP in vehicle vs. testosterone conditions (prepairing [lower 95% CI

=

0.9508, higher 95% CI

=

1.084] vs. postpairing [lower 95% CI

=

1.274, higher

95% CI

=

1.786] in vehicle conditions,

=

5 cells from three mice, two-tailed paired

test,

=

0.0038 [observed power

=

0.992, Cohen

’

=

2.704, difference

between means

=

0.5124

±

0.0847, 95% CI

=

0.2771 to 0.7478], prepairing [lower 95% CI

=

0.9349, higher 95% CI

=

1.120] vs. postpairing [lower 95% CI

=

1.456,

higher 95% CI

=

3.334] in testosterone conditions,

=

4 cells from three mice, two-tailed paired

test,

=

0.0209 [observed power

=

0.889, Cohen

’

=

2.232,

difference between means

=

1.368

±

0.3063, 95% CI

=

0.3926 to 2.342], postpairing in vehicle [lower 95% CI

=

1.274, higher 95% CI

=

1.786] vs. testoster-

one [lower 95% CI

=

1.456, higher 95% CI

=

3.334] conditions,

=

–

5 cells from six mice, two-tailed unpaired

test,

=

0.0174 [observed power

=

0.932, Cohen

’

=

0.6388, difference between means

=

0.8652

±

0.2974, 95% CI

=

0.2044 to 1.526]). (

) Schematic of the experimental design used to trigger and record

behaviorally induced LTP in vivo in NONs. (

)

fEPSP amplitude over time, prior to and following social behavior in the resident

–

intruder assay, in vehicle- vs.

testosterone-treated NON mice (average fEPSP from

=

3micepergroup).(

) Average fEPSP amplitude immediately prior to and following the expression of

social behavior in the resident

–

intruder assay, in vehicle- vs. testosterone-treated NON mice. (S) Quantification of fEPSP amplitude, prior to and following

the induction of LTP in vehicle vs. testosterone conditions (prepairing [lower 95% CI

=

1.027, higher 95% CI

=

1.123] vs. postpairing [lower 95% CI

=

0.7907,

higher 95% CI

=

1.143] in vehicle conditions,

=

3 mice, two-tailed paired

test,

=

0.1020 [observed power

=

0.999, Cohen

’

=

1.6667, difference between

means

=

0.1081

±

0.0374, 95% CI

=

−

0.2692 to 0.05303], prepairing [lower 95% CI

=

0.7055, higher 95% CI

=

1.209] vs. postpairing [lower 95% CI

=

1.027, higher

95% CI

=

2.012] in testosterone conditions,

=

3 mice, two-tailed paired

test,

=

0.0098 [observed power

=

0.786, Cohen

’

=

5.7787, difference between

means

=

0.5625

±

0.0562, 95% CI

=

0.3207 to 0.8043]). (

) Representative behavior raster plot of the same mouse treated with vehicle and 8 d after with tes-

tosterone and used for in vivo electrophysiology experiments. ns, not significant; *

<

0.05, **

<

0.01, ****

<

0.0001. In box plots, the median is represented by

the center line, the interquartile range is represented by the box edges, the bottom whisker extends to the minimal value, and the top whisker extends t

othe

maximal value. In bar graphs, data are expressed as mean

±

SEM.

Stagkourakis et al.

PNAS

October 13, 2020

vol. 117

no. 41

25797

NEUROSCIENCE

Downloaded at California Institute of Technology on October 13, 2020

well as the capacity to express strong LTP at AHiPM

→

VMHvl

Esr1

synapses (both ex vivo and in vivo) following aggression training. This

observation further strengthens t

he correlation between the ability to

respond positively to aggr

ession training, and the expression of LTP.

However, it does not distinguish whether the enhanced LTP in NON

mice is directly caused by T treatment, or rather is an indirect

effect of the increased aggression promoted by the hormone

implants. A schematic summary of the findings is presented in

SI Appendix

,Fig.S5

Discussion

Synaptic plasticity mechanisms have been investigated predom-

inantly in hippocampal circuits that mediate spatial learning (48)

or in thalamo-amygdalar circuits that mediate Pavlovian asso-

ciative conditioning (49). Both systems emphasize the role of

LTP in allowing flexible neural circuits to mediate adaptive re-

sponses on fast timescales, as expected for the recently evolved

brain regions in which they operate. By contrast, studies of the

hypothalamus have emphasized circuits that mediate innate,

evolutionarily ancient survival behaviors, with the expectation

that they would be comprised predominantly of relatively stable,

hard-wired connections (50

–

52). The present results demon-

strate that LTP and LTD can occur at a hypothalamic synapse

important for aggression (27, 28) and that LTP is part of the

causal mechanisms through which repeated successful aggressive

encounters increase aggressiveness. Moreover, they suggest that

individual variation in T levels, taken together with an effect of T

to promote LTP at AHiPM

→

VMHvl

Esr1

synapses, may account

for differences among genetically identical mice in the ability to

respond to aggression training.

The form of hypothalamic LTP described here has subtle but

interesting differences from that described in the hippocampus

(see extended discussion in

SI Appendix

). Further investigation

of synaptic plasticity mechanisms within neural pathways that

control evolutionarily selected, robust survival behaviors may

yield new insights into both the physiological and hormonal

regulation of such mechanisms, as well as the forms of behavioral

plasticity that they ultimately subserve.

Materials and Methods

All experimental procedures involving the use of live animals or their tissues

were carried out in accordance with the NIH guidelines and approved by the

Institutional Animal Care and Use Committee and the Institutional Biosafety

Committee at the California Institute of Technology (Caltech).

Esr1

Cre/

knock-

in mice (22) backcrossed into the C57BL/6N background (

>

N10) were bred at

Caltech. The

Esr1

Cre/

knock-in mouse line is available from The Jackson

Laboratory (stock no. 017911). Heterozygous

Esr1

Cre/

mice were used for all

experiments and were genotyped by PCR analysis of tail DNA. Mice used as

residents (see 5cdRI assay) were individually housed. All wild-type mice used

as intruders in resident

–

intruder assays and for behavioral experiments were

of the BALB/cAnNCrl or Crl:CD1 (ICR) strain (Charles River Laboratories).

Health status was normal for all animals. Antibodies, compounds, and the

experimental procedures with the coordinates of all injection sites are de-

scribed in

SI Appendix

Data Availability.

All data discussed in the paper are available in the main text

and

SI Appendix

. We used standard MATLAB functions and publicly avail-

able software indicated in the manuscript for analysis.

ACKNOWLEDGMENTS.

We thank Dr. B. Weissbourd and Dr. L. Li for advice

on experiments, Dr. Y. Ouadah and Dr. P. Williams for advice during writing

of the manuscript, X. Da and J. S. Chang for technical assistance, X. Da and

C. Chiu for laboratory management, and G. Mancuso for administrative

support. Members of the D.J.A. laboratory are thanked for discussion during

the preparation of this manuscript. Confocal imaging was performed in the

Biological Imaging Facility, with the support of the Caltech Beckman

Institute and the Arnold and Mabel Beckman Foundation. This study was

supported by NIH Grant R01 MH070053 (to D.J.A.) and the European Molecular

Biology Organization Advanced Long-Term Fellowship 736-2018 (to S.S.). D.J.A.

is an investigator of the HHMI.

1. D. S. Manoli, G. W. Meissner, B. S. Baker, Blueprints for behavior: Genetic specification

of neural circuitry for innate behaviors.

Trends Neurosci.

, 444

–

451 (2006).

2. K. K. Ishii, K. Touhara, Neural circuits regulating sexual behaviors via the olfactory

system in mice.

Neurosci. Res.

140

,59

–

76 (2019).

3. E. B. Anderson, I. Grossrubatscher, L. Frank, Dynamic hippocampal circuits support

learning- and memory-guided behaviors.

Cold Spring Harb. Symp. Quant. Biol.

–

58 (2014).

4. E. Yavas, S. Gonzalez, M. S. Fanselow, Interactions between the hippocampus, pre-

frontal cortex, and amygdala support complex learning and memory.

F1000 Res.

–

6 (2019).

5. S. M. Sternson, Hypothalamic survival circuits: Blueprints for purposive behaviors.

Neuron

, 810

–

824 (2013).

6. J. M. Moscarello, S. Maren, Flexibility in the face of fear: Hippocampal-prefrontal

regulation of fear and avoidance.

Curr. Opin. Behav. Sci.

,44

–

49 (2018).

7. K. Miyamichi

et al

., Cortical representations of olfactory input by trans-synaptic

tracing.

Nature

472

, 191

–

196 (2011).

8. D. L. Sosulski, M. L. Bloom, T. Cutforth, R. Axel, S. R. Datta, Distinct representations of

olfactory information in different cortical centres.

Nature

472

, 213

–

216 (2011).

9. C. M. Root, C. A. Denny, R. Hen, R. Axel, The participation of cortical amygdala in

innate, odour-driven behaviour.

Nature

515

, 269

–

273 (2014).

10. H. Amin, A. C. Lin, Neuronal mechanisms underlying innate and learned olfactory

processing in Drosophila.

Curr. Opin. Insect Sci.

–

17 (2019).

11. J. B. Rosen, The neurobiology of conditioned and unconditioned fear: A neuro-

behavioral system analysis of the amygdala.

Behav. Cogn. Neurosci. Rev.

,23

–

(2004).

12. C. T. Gross, N. S. Canteras, The many paths to fear.

Nat. Rev. Neurosci.

, 651

–

658

(2012).

13. F. Gore

et al

., Neural representations of unconditioned stimuli in basolateral amyg-

dala mediate innate and learned responses.

Cell

162

, 134

–

145 (2015).

14. P. Namburi

et al

., A circuit mechanism for differentiating positive and negative as-

sociations.

Nature

520

675

–

678 (2015).

15. N. Tinbergen,

The Study of Instinct

, (Clarendon Press, Oxford, England, 1951), p. 228.

16. K. Lorenz,

On Aggression

, (Harcourt, Brace & World, Inc., New York, 1963).

17. K. M. J. Lagerspetz, S. Garattini, E. B. Sigg, Aggression and aggressiveness in labo-

ratory mice.

Aggress. Behav.

,77

–

85 (1969).

18. J. P. Scott, E. Fredericson, The causes of fighting in mice and rats.

Physiol. Zool.

273

–

309 (1951).

19. S. Stagkourakis

et al

., A neural network for intermale aggression to establish social

hierarchy.

Nat. Neurosci.

, 834

–

842 (2018).

20. S. A. Golden

et al

., Basal forebrain projections to the lateral habenula modulate

aggression reward.

Nature

534

, 688

–

692 (2016).

21. D. Lin

et al

., Functional identification of an aggression locus in the mouse hypo-

thalamus.

Nature

470

, 221

–

226 (2011).

22. H. Lee

et al

., Scalable control of mounting and attack by Esr1+ neurons in the ven-

tromedial hypothalamus.

Nature

509

, 627

–

632 (2014).

23. C. F. Yang

et al

., Sexually dimorphic neurons in the ventromedial hypothala-

mus govern mating in both sexes and aggression in males.

Cell

153

,896

–

909

(2013).

24. T. Yang

et al

., Social control of hypothalamus-mediated male aggression.

Neuron

955

–

970.e4 (2017).

25. H. Alle, P. Jonas, J. R. Geiger, PTP and LTP at a hippocampal mossy fiber-interneuron

synapse.

Proc. Natl. Acad. Sci. U.S.A.

, 14708

–

14713 (2001).

26. L. Lo

et al

., Connectional architecture of a mouse hypothalamic circuit node con-

trolling social behavior.

Proc. Natl. Acad. Sci. U.S.A.

116

, 7503

–

7512 (2019).

27. X. Zha

et al

., VMHvl-projecting Vglut1+ neurons in the posterior amygdala gate

territorial aggression.

Cell Rep.

, 107517 (2020).

28.

T. Yamaguchi

et al

., Posterior amygdala regulates sexual and aggressive behaviors in

male mice.

Nat. Neurosci.

, 1111

–

1124 (2020).

29. C. I. Myme, K. Sugino, G. G. Turrigiano, S. B. Nelson, The NMDA-to-AMPA ratio at

synapses onto layer 2/3 pyramidal neurons is conserved across prefrontal and visual

cortices.

J. Neurophysiol.

, 771

–

779 (2003).

30. J. A. Kauer, R. C. Malenka, Synaptic plasticity and addiction.

Nat. Rev. Neurosci.

844

–

858 (2007).

31.M.A.Ungless,J.L.Whistler,R.C.Mal

enka, A. Bonci, Single cocaine exposure

in vivo induces long-term potentiation in dopamine neurons.

Nature

411

,583

–

587

(2001).

32. V. Jeevakumar, S. Kroener, Ketamine administration during the second postnatal

week alters synaptic properties of fast-spiking interneurons in the medial prefrontal

cortex of adult mice.

Cereb. Cortex

, 1117

–

1129 (2016).

33. E. Harde

et al

., EphrinB2 regulates VEGFR2 during dendritogenesis and hippocampal

circuitry development.

eLife

, e49819 (2019).

34. L. Scipioni, L. Lanzanó, A. Diaspro, E. Gratton, Comprehensive correlation analysis for

super-resolution dynamic fingerprinting of cellular compartments using the Zeiss

Airyscan detector.

Nat. Commun.

, 5120 (2018).

35. R. C. Malenka, R. A. Nicoll, Learning and memory. Never fear, LTP is hear.

Nature

390

552

–

553 (1997).

36. Y. Rao

et al

., Repeated in vivo exposure of cocaine induces long-lasting synaptic

plasticity in hypocretin/orexin-producing neurons in the lateral hypothalamus in mice.

J. Physiol.

591

, 1951

–

1966 (2013).

25798

www.pnas.org/cgi/doi/10.1073/pnas.2011782117

Stagkourakis et al.

Downloaded at California Institute of Technology on October 13, 2020

37. D. Jaffe, D. Johnston, Induction of long-term potentiation at hippocampal mossy-

fiber synapses follows a Hebbian rule.

J. Neurophysiol.

, 948

–

960 (1990).

38. N. C. Klapoetke

et al

., Independent optical excitation of distinct neural populations.

Nat. Methods

, 338

–

346 (2014).

39. X. I. Salinas-Hernández

et al

., Dopamine neurons drive fear extinction learning

by signaling the omission of expected aversive outcomes.

eLife

, e38818

(2018).

40. S. Frey, J. Bergado-Rosado, T. Seidenbecher, H. C. Pape, J. U. Frey, Reinforcement of

early long-term potentiation (early-LTP) in dentate gyrus by stimulation of the ba-

solateral amygdala: Heterosynaptic induction mechanisms of late-LTP.

J. Neurosci.

3697

–

3703 (2001).

41. S. Nabavi

et al

., Engineering a memory with LTD and LTP.

Nature

511

, 348

–

352 (2014).

42. D. Natarajan, H. de Vries, D. J. Saaltink, S. F. de Boer, J. M. Koolhaas, Delineation of

violence from functional aggression in mice: An ethological approach.

Behav. Genet.

,73

–

90 (2009).

43. S. F. Anestis, Testosterone in juvenile and adolescent male chimpanzees (Pan trog-

lodytes): Effects of dominance rank, aggression, and behavioral style.

Am. J. Phys.

Anthropol.

130

, 536

–

545 (2006).

44. T. O. Oyegbile, C. A. Marler, Winning fights elevates testosterone levels in California

mice and enhances future ability to win fights.

Horm. Behav.

, 259

–

267 (2005).

45. M. S. Barkley, B. D. Goldman, Testosterone-induced aggression in adult female mice.

Horm. Behav.

,76

–

84 (1977).

46. M. S. Barkley, B. D. Goldman, The effects of castration and Silastic implants of

testosterone on intermale aggression in the mouse.

Horm. Behav.

,32

–

(1977).

47. J. Archer, Testosterone and human aggression: An evaluation of the challenge hy-

pothesis.

Neurosci. Biobehav. Rev.

, 319

–

345 (2006).

48. H. Eichenbaum, Spatial learning. The LTP-memory connection.

Nature

378

, 131

–

132

(1995).

49. H. C. Bergstrom, The neurocircuitry of remote cued fear memory.

Neurosci. Biobehav.

Rev.

, 409

–

417 (2016).

50. J. Kohl

et al

., Functional circuit architecture underlying parental behaviour.

Nature

556

326

–

331 (2018).

51. J. R. Moffitt

et al

., Molecular, spatial, and functional single-cell profiling of the hy-

pothalamic preoptic region.

Science

362

, eaau5324 (2018).

52. K. K. Ishii

et al

., A labeled-line neural circuit for pheromone-mediated sexual be-

haviors in mice.

Neuron

, 123

–

137.e8 (2017).

Stagkourakis et al.

PNAS

October 13, 2020

vol. 117

no. 41

25799

NEUROSCIENCE

Downloaded at California Institute of Technology on October 13, 2020