of 27
Neural encoding of socially adjusted value during competitive and hazardous
foraging
Brian Silston
1
* Toby Wise
2,3,4
*, Song Qi
2,
,
Xin Sui
2
, Peter D ayan
5,6
, and De an Mobbs
2,7,8
1
Columbia University, De partmen t of Ps ychology, 370 Sch erme rhor n Hall 1190 Amsterdam
Ave., N ew York, NY 1002 7, USA;
2
Departme nt of Humanities and Social S ciences and Cali fornia Insti tute of T echn ology, 120 0 E
California Blvd, HSS 228–77 , Pasadena, California 9112 5, USA.
3
Wellcome Centr e for Human Neuroimaging, Univ ersity Colleg e London, London, UK
4
Max Planck UCL Cent re for Computational Psychiatr y and Ag ein g Resea rch, Uni versity Colleg e
London, London, UK
5
Max Planck Institute for Biological Cybe rnetics, Tübing en, G ermany
6
Unive rsity of Tübingen , Tübinge n Germ any
7
Computation and Neural Systems Prog r am at the Californ ia Institut e of Tec hnol ogy, 12 00 E
California Blvd, HSS 228–77 , Pasadena
*equal contribution
8
Corresponding author:
Dmobbs@caltech.edu
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
In group foraging organisms, optimizing the conflicting demands of competitive food loss
and
safety
is
cr itica l.
We
demonstrat e
that
hum ans
selec t
competition
av oidant
and
ris k
diluting
str ategies
du ring
foraging
depending
on
socially
adjuste d
val ue.
We
formulat e
a
mathem atica ll y grounded quan tifica ti on of socially adjusted v al ue in for aging environments
and show using multivar iat e fMRI analyses that sociall y adjusted val ue is encoded by mid-
cingulate
and
ventromedi al
prefront al
cortices,
regions
that
integrate
value
and
action
signals.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
3
Across
phyloge ny,
foragi ng
decisions
( e.g.
patch
sel ection,
feedi ng
b ehavior
and
duration)
ar e
strongly in fluenced by competi tor de nsit y, food quantit y and exp ected en er gy cost
1,2
. In p redation
fre e, y et competi tive , en vironmen ts, av oiding competition de nse pa tches is a n adaptive st rate gy
to maximize gain (e. g. see
3,4
) . In contra st, foragin g decisions under poten tial t h reat of p redation
are
gover ned
by
risk
dilution
strat egi es
(i. e.
saf ety
in
numbers)
for
whic h
larg er
g roups
of
conspecifics reduce the chanc e a particu lar individual will fa ll victim to lethal at tack by predators
5-
7
. Howeve r, risk dilution st rateg ies i ncur ef ficiency costs, reducing exploitat ion a nd harv est rates
1
.
Accordingly, optimal fora ging th eories
1,2
suggest that th e conflictin g trad e-off betwe en t hr eat of
competition and the
th reat of
predatio n are miti gated by t he ove rall f itness or socially adjusted
value of the patc h. Socially adjusted v alue, thus, is r epr esent ed as t he overa ll poten tial be nef it,
whe th er in t he safe domain via selectio n of less competition dense patches or risk dilution in the
thr eat
domain
via
occupation
of
more
competition
dense
patches
and
is
de coupled
from
th e
observed
social
densit y
of
a
patch.
It
is
unclear
w he the r
humans
obey
th ese
r ules
and
whe th er
socially
adjusted
value,
independen t
of
th e
observable
statistics
of
a
fora gin g
envi ronment,
is
rep resen ted in th e human brain.
A grow ing body of l ite rature is be gin nin g to elucidate th e unde rlyin g neurobiological mecha nisms
of
foragin g
decision-making,
albei t
in
t he
absence
of
t hr eat.
Human
and
non-human
primate
researc h
has
focused
primaril y
on
virt ual
two-patch
fora gin g
tasks,
consistently
hi gh lig htin g
regions
i nvolved
in
action
select ion
an d
value
encoding
( e.g .
mid-cingulat e
c ortex
[MCC]
and
vent romedial pre frontal cortex [ vmPFC])
8,9
,
10
. Furt he r, ante rior to th e MCC is th e dorsal cingulate
cortex (dACC),
whic h
has be en linked t o foragi ng decisions and
th e di fficulty
of such decisions
(see for debate
8,9
), w hile t he vmPFC h as been linked to re pr esentations of c h oice value during
foragin g
tasks
9,11
,
w hich
is
in
l ine
wit h
its
role
i n
monitorin g
action
va lue,
exploration
an d
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
4
economic
decisions
12,13
.
Given
th e
fu nctional
he terog ene ity
of
the
MCC
a nd
vmPFC,
one
hypot hesis is that th ese r egions re flect the socially adjusted value of foraging d ecisions through
the
int egra tion
of
know ledg e
about
competition
and
th reat,
constructing
a
primary
decision
variable more immediatel y rel evant to behavior t han t he simple social density of an environmen t.
We
addressed
this
idea
by
c reati ng
f oraging
e nvironmen ts
t hat
are
ide ntic al
exce pt
for
th e
socially
adjusted
value
strat egi es.
T hus,
testin g
conditions
for
both
safe
and
ha zardous
environme nts in w hich th e optimal stra tegi es are i nve rsely cor relat ed, for example competitio n
avoidance
and
risk
dilution,
allows
us
to
investiga te
t he
ove rall
socially
adjusted
value
of
t he
patch i ndepe ndent of its dir ectly observ able statistics.
In this study, pa rticipan ts we re scanned for app roximately four hours each ove r th e course of
two
days
whil e
th ey
per formed
a
two- patch
fora gin g
task
wit h
chang ing
l ev e l
of
thr eat
and
competition density. W e examined multivariate , distributed neura l re pr esentat i ons involved in
competition avoidance and risk dilution during a vi rtual foragi ng pa radigm i n whi ch pa rtici pants
assessed
competition
density
and
risk
of
pr edation
and
made
choices
to
ent er
on e
of
two
patches. First, pa rtici pants lear ned th e sequence of competitor states (sp ecificall y, the numbe r
of competitors in each patc h for re peati ng pairs of side-b y-side patch es ), and t hen w er e asked
to select th e pa tch i n whic h t he y would l ike to fora ge. Simply select ing a pa tch was insuffici en t
to rec eive a food re ward token. Food tokens app eared at random times and l ocations in t h e
patch,
and
the
play er
was
r equired
to
navigat e
to
it
be fore
othe r
competitor s
in
th e
patc h .
Ther efor e, t he hig he r th e densit y of competitors, th e less likel y it was that th e subject would be
able to acquire th e token. How eve r, in s ome patches (ide ntif ied by t he color), a predator could
appea r
randomly
at
any
time
or
locati on
and
would
captu re
th e
playe r
or
o ne
of
t he
othe r
competitors, also at random.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
5
We hy pothesiz ed that subjects would adapt the ir
foragin g decisions based on the unde rlyin g
socially
adjusted
value
of
a
patch,
bei ng
hig he r
in
lo w
social
densi ty
patch es
during
safe
foragin g
(as
a
result
of
hi gh
re ward
av ailability),
and
hig he r
in
more
socially
dense
patch es
during
t hr eat
of
p redation
(as
a
resu lt
of
d ecreased
risk
of
p redation).
W e
additionally
hypot hesized that th ese decisions would be supported by n eural e ncoding of s ocially adjusted
value indep enden tly of social density pe r se. Importan tly, our saf e and haza rdous patches w er e
matched for t he ef fort of d ecision, en er gy costs, and competition and rewa rd d ensity. Fur the r,
patch switc h costs wer e ze ro, the re fore allowing us to investi gate pure context ual changes i n
socially adjusted value of the decision.
We assessed decision making by condit ion by computing t he p erce ntag e of t ria ls in w hich t he
playe r
select ed
th e
patch
wit h
f ew er
c ompetitors
(Fig.
1,
pan el
E.),
and
in
more
fi ne
g rain ed
detail
by
calculating
a
dif fer ence
score
refl ectin g
th e
number
of
competitors
p resent
in
each
patch at t he time of decision. Par ticipan ts selected th e less populated patch in 8 9% of decisions
for
the
saf e
condition
and
in
32%
of
decisions
in
the
th reat
condition,
inclu sive
of
all
trial
durations
(
χ
2
=
4046,
p<.0 005,
p roport ion;
t(2 1)
=
9.07 ;
p <.0 005,
total
count).
Furth er,
we
observed
dif fe rent
av erag e
diff er ence
scores
for
saf e
a nd
t hr eat
conditions
(t (24)
=
9.14,
p<.0 005 ), indicatin g a context dep ende nt sh ift in be havioral decision-making b ased on socially
adjusted value of each pa tch.
The
for egoin g
tend ency
to
congr egat e
in
numbers
in
t he
pres ence
of
pot e ntial
p redation
optimized t he probability of avoiding ca pture, ev en for t he smallest patc h dif fe r ence (Figure 1,
panel F). The conve rse was t rue for th e s afe cont ext, such that avoiding competit ion result ed i n
the maximizing token collection (Fi gure 1, panel D). D ecisions to forage in patc hes wit h fe we r
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
6
competitors
incr eased
capture
p robability
compar ed
wit h
foragi ng
i n
pa tch es
wit h
more
competitors, ev en for t he closest value diffe rences, e. g. be twe en havin g 2 mor e pla yers in th e
patch or two fe we r play ers in t he patc h (
χ
2
=109.9, p <.0 005 ). Playe rs also collected significant l y
few er
food
tokens
in
the
t hrea t
than
the
safe
condition,
likely
re flecti ng
both
increased
competition
and
distraction
due
to
anticipation
of
the
pr edator
(t(38 )
=
5.93,
p<.0 005 ;
Fig
1,
panel D). To assess th e eff ect of distrac tion of thr eat above and b eyond basic c ompetition we
assessed
mean
token
collection
at
each
level
of
competit ion
across
both
safe
and
thr eat
conditions, and overall across conditions (see Fi g S2 in supplem entar y mate rials). ]
Optimization i n t he safe domain via sel ection of less compet itiv e patch es h eld across all four
blocks
and
began
ear ly
i n
t he
fi rst
bl ock,
refl ecting
spontaneous
and
sw ift
acquisition
of
adaptive decision-making be havior. Lik ewise, most, but not all participa nts quickly adopted a
risk dilution strategy i n th e pr esence of predation, e videnced by consistent de cision behavior
across
all
four
trial
blocks.
While
some
partici pants
adopted
an
ident ical
strate gy
across
trial
type ( e.g. saf e / th reat ), most decisions i n safe t rials re flect ed a competition avoi dance strat egy,
that
is,
to
a
patc h
wit h
f ew er
competi t ors
in
order
to
maximize
gain
(see
Fi gure
1,
pane l
A),
whil e th e op posite was t rue for th reat t ri als, on av erag e, suggesti ng a risk dilutio n strat eg y (see
Figure 1, pan el B).
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
7
Figure 1
A:
Task
Design.
In
the
safe
pla y
p hase
the
play er
selects
th e
desi red
patch
a n d
the n
co m petes
to
capture
re wards;
B:
i n
th e
danger
phase
t he
p laye r
selects
t he
des ired
patc h
and
then
co m petes
fo r
r ewa rds,
and
is
subject
to
po ten tial
captu re
by
th e
p redat o r.
In
t he
saf e
co nditio n the trial ends af ter tim e expi re s. The side -by-side patch es we re th e sa m e co lo r during
the ex per im ent and a re just co lo r-co ded h er e to clari fy th e diff er ences b etw een safe and t hrea t
co nditio ns.
C:
The
risk
calculus
that
i nfo rm s
individual
decisio n
m aking
based
o n
so cially
adjusted
value.
In
safe
patch es
(blue
lin e)
t he
o p tim al
strat eg y
is
to
sel ect
t he
patch
wit h
t he
few est
num ber
o f
co m pet ito rs,
t hus
m axim izing
re ward
gain
and
so cially
adj usted
value.
In
dangero us
patches
(black
line )
incr easing
gro up
size
t hr eate ns
the
abili ty
to
capture
r ewa rds
but
dilutes
risk
o f
bei ng
the
tar get
o f
t he
pr edato r.
d
o
r ep resents
a
decisio n
in
wh ich
so ciall y
adjusted
value
has
be en
m axim ized
in
th e
saf e
co nditio n;
d
1
re prese nts
an
i ndividual
wit h
m o derate
risk
to le rance
i n
t he
dang er
co nditio n,
willin g
to
sel ect
a
patc h
wi th
se ve ral
co m petito rs
in
o rde r
to
r educe
captur e
risk
w hil e
sti ll
co m peting
fo r
re war ds;
D:
Pla yer s
co llected
signi ficantly
m o re
r ewa rds
in
safe
t han
in
th reat
patc h
co nfigu rati o ns;
E:
Players
spo ntaneo usly ado pted a co m petitio n a vo idance strat egy in t he safe co nditio n, whil e m o st but
no t all play ers ado pted a risk dilutio n str ategy in t he t hr eat co nditio n; F: P ro bability distributio n
o f
being
captured
as
a
functio n
o f
th e
diffe renc e
sco re
o n
a
given
t rial.
A
large
po sitive
diffe rence
sco re
indicates
pres ence
in
a
patch
wit h
f ew
o r
o n e
o the r
co m petito r.
A
la rg e
negati ve sco re indicates p rese nce in a pa tch wi th se ve ral o th er co m pet ito rs.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
8
In order to p rovide a mathematicall y g rounded quantification of socially adjusted value, th e ke y
decision variable emer ging in our behav i oral analyses, we fit a model to subjects' decisions t hat
made
decisions
based
on
the
socially
adjusted
value
of
each
patch
(see
Met hods).
Socially
adjusted
value
depended
on
th e
av era ge
number
of
points
coll ected
in
each
condition,
and
included a free pa ramete r re pr esenti ng the value of r eceiv ing a shock, whic h w as negative for
all
but
one
subject
(Figure
2A).
As
exp ected,
inf er red
socially
adjusted
value
from
the
model
decreased
wit h
more
competitors
durin g
safety
and
inc reased
during
t hr eat
(F igure
2B ),
and
the
diff er ence
in
socially
adjusted
value
bet we en
t he
two
patch es
was
strong ly
p redicti ve
of
choice
(Figure
2C ),
correctl y
pr edicting
66.72%
of
choices.
Predicted
probabil ities
based
on
socially
adjusted
value
differ ence
w e re
also
well
calibrat ed
wit h
resp ect
to
true
choice
probabilities (Fi gure 2D).
We
next
sought
to
det ermin e
how
d ecision
variables
ar e
encoded
in
t he
brain.
We
used
rep resen tational
similari ty
analysis
(RS A)
to
identif y
re gions
w her e
activi ty
p atterns
ali gne d
with
th e
task
structur e
durin g
t he
deci sion-making
phase
of
t he
task,
during
whic h
subjects
wer e
awar e
of
available
options
but
could
not
yet
make
a
motor
response.
Critically,
RSA
allowed
us
to
ident ify
multivariat e
re prese ntations
of
ke y
decision
va riable s,
rath er
than
assessing
chang es
i n
sin gle -voxel
acti vi ty
l evels
as
in
t raditional
uni variat e
an alyses,
t hrough
identif ying
r egions
wh er e
n eural
simil arity
across
conditions
aligns
wit h
si milarity
in
th e
prope rti es
of
the
task
being
p erfo rmed.
While
univa riate
anal yses
could
only
dete rmin e
whe th er
socially
adjusted
value
is
associated
wit h
activ ity
in
eac h
r egion,
RS A
allows
us
to
determin e
wh eth er
t he
multivariat e
re prese ntation
of
socially
adjusted
value
in
a
reg ion
is
consistent
across
dif fer ent
conditions.
We
computed
rep rese ntational
dissimilarity
matrice s
(RDMs)
for
BOLD
responses
to
each
t ri al
and
used
linear
re gression
to
pr edict
the
observed
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
9
patte rn of neural similarit y f rom RDMs f or task conditions (Figure 2D). First, w e computed task
RDMs based socially adjusted value (as determin ed using our be havioral model) in th e curr en t
patch
( th e
patc h
select ed
on
t he
pre vi ous
trial)
and
th e
alt ernat ive.
Second,
we
computed
RDMs
rep rese ntin g
th e
socially
adjusted
value
of
the
curr ent
and
alt erna tive
patch.
We
also
included RDMs repr esenti ng t he dif fer e nce in competitors and socially adjusted value betwe en
patches and i ncluded an RDM re pr esenti ng t he eff ect of th reat.
We used a searchligh t app roach, pe rfor ming the RSA analysis in with in 6mm spher es across the
brain
to
localize
r egions
encoding
task- rele vant
r epr esenta tions.
Th is
ide ntif ie d
a
distributed
network
of
re gions
e ncoding
t he
socially
adjusted
value
of
th e
al ter nativ e
patc h
durin g
decision-making, including th e MCC, posterior cingulat e cortex (PCC), medial p r efrontal cort ex
(mPFC), and orbitofrontal cort ex (OFC) ( voxel ps < . 05, th reshold-f re e cluster cor rection, Fi gur e
2E). The socially adjusted value of the c urrent patch was also repr esented in th ese re gions, but
was additionally r ep resent ed i n th e pre motor cortex, hip pocampus, and ant eri or insular corte x
(voxel ps < .05, t hr eshold-f re e cluster correct ion, Figure 2E). Importa ntly, our an alysis approach
isolated
unique
eff ects
of
each
task
RDM,
controllin g
for
e ff ects
of
othe r
task
RDMs.
In
contrast, th e number of competitors i n th e alt er nativ e pa tch was only r ep rese nted in t he le ft
pre -suppleme ntar y motor ar ea (voxel p s < .0 5, t hr eshold-f re e cluster correct ion), and no are a
rep resen ted
t he
number
of
competitor s
in
the
curr ent
patch.
Notably,
w e
fo und
no
region
encoding t he di ff ere nce bet we en patch es, eit he r in te rms of number of compet itors or socially
adjusted value. Threat
lev el (i .e. saf e or at risk of p redation)
was encoded in a
wide ra nge of
cortical and subcortical reg ions (se e Fi g ure S3), including t he MCC, vmPFC, hip pocampus and
amygdala.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
10
Figure 2
A:
Values
o f
the
sho ck
co st
param eter
f ro m
o ur
behavio ral
m o del,
indicating
th at
sho cks
wer e
perceived as a cost for all but one subject. B: Socially adjusted value across task conditions,
dem o nstrating
that
so cially
adjusted
v alue
dep ends
o n
bo th
th reat
l evel
and
th e
num ber
o f
co m petito rs. C) Th e pro bability o f cho o sing a patc h based o n th e di ff ere nce b etw een its so ciall y
adjusted
value
and
that
o f
the
alte rnati ve
patch .
The
functio n
sho wn
r ep resent s
the
results
o f
lo gistic
re gressio n
m o dels
predicti ng
cho ice
fro m
so cially
adjusted
value
diffe re nce
(se e
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
11
methods).
D)
Calibra tion
plot
sho win g
predict ed
probability
of
c hoice
bas ed
on
socially
adjusted
value
differenc e
deriv ed
fro m
a
logistic
regression
model
versus
the
t rue
choice
probability for safe and th reat condition s. Probabilities are binn ed i nto 10% bins, and er ror ba rs
rep resen t 95 % confide nce i nte rvals acr oss subjects. E: Rep resenta tional dissimilarity matrice s
(RDMs) for th e n eural data
(hy poth etic al example s hown l eft ) and task conditi ons of inte rest
(rig ht) .
Neural
RDMs
w er e
modeled
as
a
function
of
task
RDMs
using
l inea r
re gression,
wit h
each RD M wei ght ed by a wei gh t param eter
β
⤂⢈⢗
. Here, RDMs are shown for a subset of trials
for a singl e subject. In the task RDMs, r ows and columns repr esent i ndividual t rials, w hile the
colors in the matrix r ep resent
th e diff er ence in socially adjusted value betwe en that tr ial and
other
tria ls.
RDMs
a re
shown
for
the
socially
adjusted
value
of
t he
cur ren t
and
alter nativ e
patch,
and
oth er
RDMs
ar e
described
t he
met hods
and
results.
F:
Left
pan el:
eff ects
of
the
current
patch
socially
adjusted
value
R DM,
showin g
widespr ead
e ffects,
inclu ding
t he
mid-
cingulate
cort ex
(MCC) ,
poste rior
cin g ulate
cortex
(PCC) ,
medial
p re frontal
cortex
(mPFC),
orbitofrontal cortex (OFC).. Ri gh t pan el: eff ects of t he al te rnativ e patc h socially adjusted value
RDM,
showin g
e ff ects
across
similar
areas,
wi th
the
most
prominent
clusters
in
the
mid
cingulate cort ex (MCC), and mPFC/OFC. Maps re pr esent p values det ermin ed using t hr eshold-
fre e cluster correction (TFCE), t hr esholded at p < .05. G) Extracted beta values from the MC C
and
vmPFC,
taken
from
the
AAL
atla s
14
mid-cingulate
and
f rontal
medial
orbital
re gions
respect ivel y, in addition to th e AAL hi pp ocampus and amygdala regions. High er values indicate
great er similari ty b etw een t he task RD M and t he n eural
RDM. T hese values a re
provided for
illustration only; signi ficance was det er mined using voxel wise t ests as shown i n (F). E rror bar s
rep resen t 95% confidence int ervals.
Our
results
demonstrate
t hat
humans
adaptively
select
social
envi ronments
based
on
thei r
socially
adjusted
value,
using
a
competition-avoidant
strate gy
wh en
t hr eat
is
absent
and
switchin g
to
a
risk-dilutin g
st rate gy
i n
the
pr esence
of
t hr eat.
Im portantl y,
t h e
key
variabl e
underpinn ing
t his
decision,
the
socially
adjusted
value
of
a
foraging
patc h,
wa s
encoded
in
a
network focused on the MCC and vmPFC, suggesting t hat th ese re gions inte gra te informatio n
about
social
density
and
threat
to
calculate
the
ove rall
socially
adjusted
value
of
both
the
current
patch and an alt ernat ive. Our r esults identify dist ributed neural syste ms repres entin g
key decision variabl es underl ying adapta tive foragin g i n r esponse to competit ion and thr eat.
Behavio rally,
w e
found
t hat
subjects
a dapted
t hei r
decision-making
strate gy
based
on
th e
socially
adjusted
value
of
foragin g
patc hes.
Und er
conditions
of
safe ty
subject s
we re
biased
towards
patch es
wi th
low
competit ion
density,
as
found
in
p re vious
work
10
,
rep resen tin g
a
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
12
competition-avoidant
strat egy .
In
cont rast,
wh en
unde r
th reat,
subjects
chose
patch es
wit h
hig h
social
d ensity,
r ep resen ting
a
ris k-dilution
strat egy.
T hese
r esults
indic ate
t hat
w hil e
foragin g in social environmen ts with t h reat of pr edation, human subjects base thei r decisions
on the socially adjusted value of available patches. In t he absence of pr edation ri sk, immediate
survival depe nds on maximizing food resources, and is there fore hig hest in envi ronments wit h
the fe west competitors. Whe n unde r t hr eat of pr edation, survi val focus shifts to risk of captu re ,
and
is
hig he r
in
env ironments
wit h
a
great er
number
of
competi tors,
and
r emains
so
until
competition
density
outwei ghs
t he
risk
dilution
bene fit.
In
real- world
envi ronments,
socially
adjusted value will depend on the i nte ra ction betwee n multiple factors, and our task presents a
simplistic case wh er e r ewa rd is r eadily a vailable. For example, wh en food is pers istently scarce,
socially adjusted value is l ikely to be re l atively h ig h in t he absence of competiti on eve n unde r
thr eat.
At the neural l ev el, w e found that th e s ocially adjusted value of a social decision envi ronment
was
encoded
across
a
distributed
netw ork
of
regions,
p rimarily
located
in
the
vmPFC,
OFC,
MCC
and
PCC.
In
contrast,
we
found
little
evidenc e
for
a
re pr esentat ion
of
t he
number
of
competitors
p er
se,
o r
t he
dif fer ence
in
socially
adjusted
value
or
social
de nsity
be twe en
patches.
Th ese
results
indicate
t hat
w hen
making
decisions
about
foraging
patches
in
t h e
prese nce
of
competitors,
t he
brain
pr e dominantly
rep rese nts
the
socially
adjusted
value
of
a
potential patch es, r egardl ess of wh et her this r ep resen ts risk of p redation or risk of competition
for food, whil e potent ially disre gardin g the observed number of competitors, or th e di ffe re nce
betwe en ava ilable patch es. Importan tly, this suggests t hat in social foragin g env ironments t h e
brain is r ep resent ing th e value of social density in te rms of a socially adjusted value, rat he r tha n
simply re prese ntin g th e numbers of competitors, and suggests t hat th e brain enc odes the value
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint
13
of independ ent options rat he r tha n th ei r diff ere nce. Our use of RSA allowed us to focus on the
multivariate re pres entations of t hese va riables in t he b rain, rath er than rel ying on single- voxel
activity chan ges. Thus, th e re gions we i dentify do not simply show similar activi ty lev els across
hig h
socially
adjusted
value
conditions
but
represen t
survival
in
t he
same
manner
across
conditions. Our focus on multivariate re prese ntations of decision variabl es during fora ging, a s
opposed to overal l activ ity lev els, distin guishes t his work f rom pr ior studies, whi ch to date hav e
exclusively
used
univariat e
app roaches.
Thus,
whil e
activit y
lev els
may
be
modulated
by
the
diffe rence
b etw een
op tions,
th e
patt ern
of
acti vity
re pres ents
th e
value
of
each
optio n
indepe ndentl y.
Concernin g socially adjusted value, two regions ar e of historical int erest
9
, th e M CC and vmPFC.
These re gions are active across both curre nt and alt er nativ e patc hes. Impor tan tly, our r esults
show tha t t hese r egions encode t he socially adjusted value of an env ironment a cross safe and
thr eat
conditions,
rat he r
t han
t he
social
density
of
competition
alon e.
W hat
are
thes e
t wo
regions
computing?
It
is
important
to
state
that
our
MCC
cluster
is
more
posterior
t han
t h e
dACC
area
that
has
been
li nked
to
the
dif ficulty
of
fora gin g
decisions
and
the
value
o f
alterna tive
options
8,10,13
.
Our
task
conditions
of
int er est
ar e
la rg ely
o rthogo nal
to
decision
difficulty.
Thus,
our
lack
of
dACC
a ctivity
may
re flect
t he
match ing
of
d ifficulty
across
conditions.
Our
results
a lso
indicate
t hat
the
MCC
is
not
purely
sig nalin g
t he
value
of
an
alterna tive op tion, or t he dif fe renc e bet wee n options, but simultaneously rep re sents both th e
value of both th e cur rent and alt erna tiv e option.
9,11,15
How eve r, based on th e kn owledge of t his
region ’s
connectivity
and
function,
w h ich
suggests
it
plays
litt le
role
in
va lue-based,
goal-
directed be havior
per se
16
, it s eems inc orrect to say t hat t he MCC re flects pure value. I nstead,
the
MCC
may
act
as
a
hub,
coordinati ng
emotional
r esponses
and
motor
actions
according
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint
this version posted September 12, 2020.
.
https://doi.org/10.1101/2020.09.11.294058
doi:
bioRxiv preprint