A icle
Decision-making abili y, psychopa hology, and b ain
connec i i y
Highligh s
dYoung people ha e a gene al decision-making abili y, which
we call ‘‘decision acui y’’
dDecision acui y is e lec ed in how s ongly connec ed ce ain
b ain ne wo ks a e
dLow decision acui y is associa ed wi h gene al social unc ion
psychopa hology
Au ho s
Michael Mou oussis, Benjamı
´n Ga zo
´n,
Sha on Neu eld, ..., NSPN Conso ium,
Ma c Gui a -Masip, Raymond J. Dolan
Co espondence
m.mou [email protected]
In b ie
Mou oussis, Ga zo
´n, e al. epo ha
young people ha e a gene al decision-
making abili y, e lec ed in he unc ional
connec i i y o speci ic b ain ne wo ks
and educed in hose wi h symp oms o
poo gene al social unc ion. Thus, i may
be impo an o unde s anding aspec s
o men al heal h and i s basis in b ain
unc ion.
Mou oussis e al., 2021, Neu on 109, 2025–2040
June 16, 2021 ª2021 The Au ho s. Published by Else ie Inc.
h ps://doi.o g/10.1016/j.neu on.2021.04.019 ll
A icle
Decision-making abili y,
psychopa hology, and b ain connec i i y
Michael Mou oussis,
1,2,7,10,
*Benjamı
´n Ga zo
´n,
3,7
Sha on Neu eld,
4
Dominik R. Bach,
1,2,5
F ancesco Rigoli,
6
Ian Goodye ,
4
Edwa d Bullmo e,
4
NSPN Conso ium, Ma c Gui a -Masip,
2,3,8,9
and Raymond J. Dolan
1,2,8,9
1
Wellcome Cen e o Human Neu oimaging, Uni e si y College London, London WC1N 3BG, UK
2
Max Planck Uni e si y College London Cen e o Compu a ional Psychia y and Ageing Resea ch, London WC1B 5EH, UK
3
Aging Resea ch Cen e, Ka olinska Ins i u e, S ockholm, Sweden
4
Depa men o Psychia y, Uni e si y o Camb idge, Camb idge CB2 0SZ, UK
5
Compu a ional Psychia y Resea ch, Depa men o Psychia y, Psycho he apy, and Psychosoma ics, Psychia ic Hospi al, Uni e si y o
Zu ich, 8032 Zu ich, Swi ze land
6
Depa men o Psychology, Ci y Uni e si y, London, UK
7
These au ho s con ibu ed equally
8
These au ho s con ibu ed equally
9
Senio au ho
10
Lead con ac
*Co espondence: m.mou ous[email p o ec ed]
h ps://doi.o g/10.1016/j.neu on.2021.04.019
SUMMARY
Decision-making is a cogni i e p ocess o cen al impo ance o he quali y o ou li es. He e, we ask whe he a
common ac o unde pins ou di e se decision-making abili ies. We ob ained 32 decision-making measu es
om 830 young people and iden i ied a common ac o ha we call ‘‘decision acui y,’’ which was dis inc
om IQ and e lec ed a gene ic decision-making abili y. Decision acui y was dec eased in hose wi h abe an
hinking and low gene al social unc ioning. C ucially, decision acui y and IQ had dissociable b ain signa u es,
in e ms o hei associa ed neu al ne wo ks o es ing-s a e unc ional connec i i y. Decision acui y was eli-
ably measu ed, and i s ela ionship wi h unc ional connec i i y was also s able when measu ed in he same
indi iduals 18 mon hs la e . Thus, ou beha io al and b ain da a iden i y a new cogni i e cons uc ha unde -
pins decision-making abili y ac oss mul iple domains. This cons uc may be impo an o unde s anding
men al heal h, pa icula ly ega ding poo social unc ion and abe an hough pa e ns.
INTRODUCTION
E ec i e decision-making unde pins a ange o ac i i ies ha
span economic pe o mance and social adap a ion. A compu a-
ional cha ac e iza ion o decision-making p ocesses is also
conside ed impo an in ad ancing an unde s anding o psychi-
a ic diso de s (Scholl and Klein-Fl€
ugge, 2018). Ye , unlike adi-
ional cogni i e cons uc s such as in elligence, he dis ibu ion
and co a ia ion o decision-making cha ac e is ics in he popu-
la ion is unknown, while he eliabili y o beha io al asks ypically
used o measu e hese abili ies has been ques ioned (B own
e al., 2020;Enka i e al., 2019;Hedge e al., 2020). Likewise,
al hough he e is a g owing knowledge ega ding he neu al un-
de pinnings o decision-making abili y, he e is a ela i e dea h
o knowledge in ela ion o adolescence and ea ly adul hood, a
c ucial pe iod o b ain ma u a ion (Giedd, 2004;Whi ake
e al., 2016). Thus, he e is an inc easing u gency in unde s and-
ing he neu al basis o cogni i e de elopmen in young people,
including i s ela ionship wi h b ain connec i i y (S ipada e al.,
2020). An added mo i a ion he e is he obse a ion ha a high
p opo ion o psychopa hology eme ges du ing adolescence
and ea ly adul hood (Paus e al., 2008).
Decision-making e lec s a complex in e play be ween mul-
iple p ocesses ha bea on e alua ing op ions and choosing
a cou se o ac ion. These p ocesses a e well cha ac e ized
wi hin a ein o cemen -lea ning amewo k (Dolan and Dayan,
2013;Kable and Glimche , 2009 Phelps e al., 2014;Su on
and Ba o, 1998). He e, a dis inc ion is made be ween a eli-
ance on lea ning how bene icial an ac ion has been in he
pas , o al e na i ely he exploi a ion o an accu a e model
o an en i onmen , in o de o in e he consequences o
each ac ion. Compu a ionally, his encompasses model- ee
con ol, accu a e model lea ning (Fehe daSil aandHa e,
2020), and model-based e alua ion (Daw e al., 2005;Dolan
and Dayan, 2013). Model-based and model- ee in luences
ade o a di e en le els in di e en indi iduals (Eppinge
e al., 2017;Kool e al., 2017).
A mo e sub le sou ce o decision a iabili y is he impac o
Pa lo ian heu is ics, e lec ing a p opensi y o a ach alue o
speci ic ac ions by me e associa ion wi h whe he hey lead o
Neu on 109, 2025–2040, June 16, 2021 ª2021 The Au ho s. Published by Else ie Inc. 2025
This is an open access a icle unde he CC BY license (h p://c ea i ecommons.o g/licenses/by/4.0/).
ll
OPEN ACCESS
ewa d o punishmen (de Boe e al., 2019;Gui a -Masip e al.,
2012;Mou oussis e al., 2018). This con lic is also e iden when
indi iduals balance a need o ha es ewa ds agains po en ial
dange s inhe en in ac ing wi hin an unce ain en i onmen
(Bach e al., 2020;Loh e al., 2017;O’Neil e al., 2015). This en-
gende s a con lic be ween mo i a ional d i es (e.g., app oach
e sus a oidance) ha need esolu ion in o de o enac e ec i e
decisions.
The e is much a iabili y in decision-making ac oss indi id-
uals. One sou ce o his a iabili y pe ains o unce ain y in
decision ou comes, whe e a ole ance o unce ain y can d i e
p e e ences o isky bu , on a e age, good op ions (Ch is opou-
los e al., 2009;Payzan-LeNes ou e al., 2013). Likewise, indi id-
ual a iabili y in decision-making is seen also in he empo al
domain, whe e indi iduals balance exploi ing an immedia ely
a ailable sa e op ion agains he possibili y o g ea e , possibly
unce ain, u u e bene i (Bad e e al., 2012;Su on and Ba o,
1998). Finally, as many decisions a e enac ed in a social con ex ,
unde s anding he in en ions and emo ions o o he s is o en
c ucial o making decisions and impac s on cha ac e is ics
such as one’s p opensi y o coope a e wi h o he s (Fe e al.,
2012;Hula e al., 2015;Luo e al., 2018).
Al hough he abo e emphasizes disc e e ac o s as in lu-
encing decision-making, we hypo hesized ha he e would
also be co a ia ion ac oss decision-making abili ies wi hin he
popula ion, implying sha ed a iance along la en dimensions.
This is analogous o he s uc u e o in elligence, whe e a co nu-
copia o abili ies co- a ies wi h la en dimensions such as gen-
e al and domain-speci ic in elligence ( an de Maas e al.,
2006). On his basis, we employed a b oad- anging decision-
making ba e y and adminis e ed i o 830 14- o 24-yea -olds
li ing in he communi y (Kiddle e al., 2018). The ba e y included
asks apping in o sensi i i y o gains and losses (mos asks in
Table 1), he ex en o which model-based in luences domina e
choice e alua ion (Table 1, ask D bu also asks C, E, and F), a
p opensi y o ake isks and exhibi impulsi i y ( asks B, C, E,
and G), and an abili y o make bene icial social judgemen s
( asks E and F). We hypo hesized ha hese ou domains would
co espond o la en dimensions o decision-making abili y
ac oss asks. We used compu a ional modeling and key desc ip-
i e s a is ics o ex ac ele an me ics om he asks (Bach
e al., 2020;Fe e al., 2012;Mou oussis e al., 2011,2016,
2018;Rigoli e al., 2016;Shaha e al., 2019a). Submi ing hese
componen me ics o ac o analysis (see STAR Me hods) al-
lowed us o de i e la en ac oss- ask cogni i e cons uc s un-
de lying decision-making and es o he p esence o la en di-
mensions co esponding o he hypo hesized cogni i e
domains.
We assessed cons uc s abili y using he da a o 571 o ou
pa icipan s who pe o med he decision-making ba e y a sec-
ond ime, a a ollow-up 18 mon hs la e on a e age, by cha ac-
e izing he ela ionship be ween he in e ed la en cogni i e
cons uc s and ex e nal measu es such as age, IQ, and men al
heal h cha ac e is ics. He e, we hypo hesized ha la en dimen-
sions o decision-making would co ela e wi h sel - epo ed psy-
chological disposi ions and men al heal h symp oms. To es his
la e hypo hesis, we a ailed pa icipan s’ de i ed sco es o
bo h gene al and speci ic disposi ion ac o s (Polek e al.,
2018) as well as concu en men al heal h symp oms (S Clai
e al., 2017).
C ucially, we cha ac e ized he neu al ci cui y unde pinning
la en decision-making ac o s. To achie e his, we analyzed
unc ional connec i i y om es ing-s a e unc ional magne ic
esonance imaging ( MRI) da a ( sFC), p o iding a me ic o
coupling be ween blood-oxygen-le el-dependen (BOLD) ime
se ies om di e en b ain egions o ne wo ks (nodes). Pa e ns
o sFC a e known o beha e as disposi ions o a la ge deg ee
(Finn e al., 2015), including p edic ing a subjec ’s cogni i e abil-
i ies in di e se domains (Dubois e al., 2018;Kong e al., 2019;
Rosenbe g e al., 2016;Smi h e al., 2015). Thus, we could ask
whe he dis inc connec i i y ne wo ks p edic ed la en deci-
sion-making ac o s and whe he iden i ied connec i i y ne -
wo ks had s abili y o e ime.
We ound e idence o a single dimension o co a ia ion in he
popula ion o which mul iple decision-making asks con ibu ed.
This dimension, which we e med ‘‘decision acui y.’’ e lec ed
speed o lea ning, an abili y o ake accoun o cogni i ely dis an
ou comes, and low decision a iabili y. We ound ha decision
acui y has a eliabili y ha was much highe han ha epo ed
o ypical decision-making asks (Mou oussis e al., 2018). In
keeping wi h his, i was associa ed wi h dis inc pa e ns o
sFC. Finally, decision acui y was cha ac e ized by a unc ional
connec i i y signa u e and a ela ionship o bo h psychological
disposi ions and symp oms ha was dis inc o ha o IQ.
RESULTS
Decision acui y is an impo an dimension o decision-
making
A o al o 830 young people aged 14–24 we e es ed using a ask
ba e y assessing di e se componen s o decision-making (Ta-
ble 1). 349 pa icipan s unde wen b ain MRI a es , on he
same day as cogni i e es ing, o assess es ing-s a e unc ional
connec i i y p o iles. Scanned pa icipan s had no his o y o
neu opsychia ic diso de and no suspec ed psychia ic diag-
nosis on SCID in e iew. 50 pa icipan s wi h DSM-5 majo
dep essi e diso de we e included in he non-scanned sample
o compa e he s uc u e o hei decision-making o he emain-
ing heal hy g oup. The STAR Me hods and supplemen al in o -
ma ion p o ide u he de ail on his subg oup.
We ex ac ed 32 decision-making measu es om he ba e y,
which we subjec ed o ac o analyses. Explo a o y ac o anal-
ysis was ollowed by con i ma o y analysis and ou -o -sample
es ing o he bes ac o model (see STAR Me hods o de ails
o he ac o -analy ic app oach, including dimensionali y es ima-
ion and s abili y analyses).
Wo king wi h he la ge , baseline sample, we disce ned ou
s able decision-making ac o s. Impo an ly, only he i s o
hese loaded on measu es om mul iple asks. We named
his ac o decision acui y, o d, as i loaded nega i ely on de-
cision a iabili y measu es, especially decision empe a u e,
and loaded posi i ely on measu es con ibu ing o p o i able
decision-making, such as low empo al discoun ing and as e
lea ning a es (Figu e 1;Table S1). Thus, pa icipan s wi h high
dhad low decision a iabili y in economic- isk, in o ma ion-
ga he ing, Go-NoGo, and Two-S ep asks. They had as
ll
OPEN ACCESS A icle
2026 Neu on 109, 2025–2040, June 16, 2021
Table 1. Decision-making ask ba e y
Task (wi h key e e ence)
B oad (selec ed)
psychological domains
Compu a ional cons uc s
assessed
Key indi idual pa ame e s
and desc ip i e measu es
A. Go-NoGo ask (Gui a -
Masip e al., 2012)
De aul (Pa lo ian)
p opensi ies o ac ion and
abili y o modi y hem
Impac o gains and losses
on choice
Pa lo ian biases (i.e.,
p opensi y o engage in
ac ion in o de o ob ain
ewa ds and o abs ain om
ac ion o a oid losses).
Rewa d sensi i i y,
equi alen o decision
empe a u e.
Ins umen al lea ning a e in
he appe i i e and a e si e
domains.
1. Pa lo ian bias.
2. and 3. Reac ion imes o
ac ion choices in he con ex
o h ea e sus oppo uni y.
4. Sensi i i y o ou comes.
5. Gene al bias o ac ion
a he han non-ac ion.
6. Mo i a ion-independen ,
‘‘i educible,’’ a iabili y in
decision-making.
7. and 8. Lea ning a es in he
appe i i e and a e si e
con ex s.
B. Economic p e e ences
ask (Symmonds e al., 2011)
(NB: adminis e ed a
baseline only)
Risk aking/impulsi i y
Impac o gains and losses
on choice
Baseline as e o gambling.
Risk a oidance (p e e ence
o ou come dis ibu ions o
low a iance).
9. O e all p e e ence o
gambling o e known
e u ns.
10. P e e ence weigh o
a iance, compa ed o he
mean, o an ou come
dis ibu ion, named
‘‘economic isk p e e ence.’’
11. E ec o ou come
dis ibu ion asymme y
(skewness) on p e e ences.
12. Sensi i i y o expec ed
alue o ou comes.
C. App oach-a oidance
con lic ask (Bach
e al., 2014)
Risk aking/impulsi i y
Impac o gains and losses
on choice
Abili y o complex planning
Willingness o expose
onesel o di e en le els o
isk o he sake o amassing
ewa ds.
Abili y o lea n abou ime-
dependen haza ds and plan
e icien senso imo o
sequences o minimize isk.
13.–15. Fac o -analy ic
sco es summa izing
a iance o e a
comp ehensi e se o
beha io al measu es in he
ask. App oxima ely
co esponding o sensi i i y
o o e all le el o h ea ,
sensi i i y o he ime
dependency o h ea , and
o e all pe o mance.
D. Two-s ep ask (Daw
e al., 2011)
Abili y o complex planning
Impac o gains and losses
on choice
S eng h o ‘‘model- ee’’
(i.e., based on di ec ly
lea ned alues o ac ions)
e sus ‘‘model-based’’ (i.e.,
explici ly es ima ing he
u u e consequences o
ac ions) decision-making.
16. Model-basedness:
endency o shi in decisions
as a consequence o a
di e en decision being mo e
ad an ageous acco ding o
he ansi ion p obabili ies
inhe en in he ask.
17. Lea ning a e.
18. Pe se e a ion endency.
19. Rewa d sensi i i y.
20. Eligibili y ace
(p opensi y o lea ning o
a ec no jus he cu en
s a e bu also o he s ela ed
o i ).
(Con inued on nex page)
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040, June 16, 2021 2027
eac ion imes and high lea ning a es in he Go-NoGo ask.
No e ha a decision empe a u e pa ame e can always be
e-w i en as he in e se o ewa d (and/o loss) sensi i i y.
Hence, he p ominen ole o nega i ely loading empe a u e
pa ame e s in dsuppo s ou ap io ihypo hesis ha ewa d
sensi i i y cons i u es an impo an sha ed cha ac e is ic
ac oss asks.
In he baseline sample, we con i med ha dco ela ed wi h
p o i able decision-making by es ima ing a measu e o agg e-
ga e ask pe o mance, based on ne poin s won ac oss asks
and sepa a e om componen s o d(Pea son = 0.50, p <
1e10; see supplemen al in o ma ion, pa C, o de ails).
Rema kably, dp edic ed his agg ega e measu e o pe o -
mance independen ly om IQ, p o iding suppo i e e idence
o con e gen alidi y wi h di ec ly measu ed ask pe o mance.
In ac , he e ec o IQ on pe o mance depended on i s sha ed
a iance wi h d( he ca ea he e being ha pe o mance in asks
and dsha e common-me hod a iance).
The o he h ee ac o s de i ed om his analysis add essed
wi hin- ask beha io a he han hypo hesized global decision-
making cons uc s and we e hus o pe iphe al in e es he e.
The second selec ed he delega ed in e - empo al discoun ing
ask (D), he hi d he in o ma ion-ga he ing ask (E), and he
ou h he economic isk p e e ence ask (C) (Figu e S2). As ex-
pec ed, gi en ha each ask had a unique ocus, cons i uen
cogni i e measu es showed high uniqueness sco es ac oss all
ac o s. 22 o he 32 measu es had uniqueness > 80%
(Figu e 1B).
Table 1. Con inued
Task (wi h key e e ence)
B oad (selec ed)
psychological domains
Compu a ional cons uc s
assessed
Key indi idual pa ame e s
and desc ip i e measu es
E. In o ma ion ga he ing ask
(Mou oussis e al., 2011)
Risk aking/impulsi i y
Abili y o complex planning
Impac o gains and losses
on choice
Assessmen o whe he
u u e decisions will be mo e
ad an ageous i one ga he s
mo e in o ma ion.
21. In o ma ion sampling
noise, which de e mines no
only decision a iabili y bu
also e ec i e dep h o
planning.
22. Subjec i e cos o e e y
piece o in o ma ion asked
o when expe imen e
imposes no such p ice
explici ly.
23. and 24. Di o i a ixed,
ex e nal p ice-pe -s ep is
imposed.
F. Mul i- ound in es o -
us ee ask (Fe e al., 2012)
Unde s anding he
p e e ences o o he s (social
cogni ion)
Abili y o complex planning
Impac o gains and losses
on choice
O e all s a egies used o
elici coope a ion and a oid
being exploi ed by one’s
anonymous ask pa ne .
25. Ini ial us (i.e., he
amoun gi en by he in es o
o he us ee be o e hey
ha e any speci ic in o ma ion
abou hem).
26. Coope a i eness:
a e age deg ee o which
in es o and us ee ended
o espond o educ ions (o
inc eases) in each o he ’s
con ibu ions in kind.
27. Responsi eness:
a e age magni ude o
esponding o he pa ne ’s
change in con ibu ion.
G. In e pe sonal-discoun ing
ask (Mou oussis e al., 2016)
Unde s anding he
p e e ences o o he s (social
cogni ion)
Risk aking/impulsi i y
Baseline in e - empo al
discoun ing; shi in
discoun ing p e e ences
upon exposu e o pee s’
p e e ences.
28. Basic hype bolic
empo al discoun ing
coe icien .
29. Rele ance o o he s’
obse ed p e e ences o he
sel .
30. Discoun ing as e
unce ain y, i.e., unce ain y
abou one’s own as es in
his domain.
31. Decision a iabili y o e
choosing o o he s.
32. I educible
decision noise.
ll
OPEN ACCESS A icle
2028 Neu on 109, 2025–2040, June 16, 2021

De elopmen al ea u es o decision acui y
We i s examined how ddepended on age, a key indica o o
de elopmen . We used linea mixed e ec s (LME) analysis wi h
pa icipan as andom e ec , wo measu emen ime poin s o
decision acui y and IQ, and one (baseline) sco e pe pa icipan
o disposi ions, sel - epo ed sex and socioeconomic a iables.
LME analysis modeled age bo h longi udinally and c oss-
sec ionally. This analysis showed ha he d a ied in he same
manne wi h age wi hin o ac oss pa icipan s (be a = 0.24,
SE = 0.022, p 0.0 [unde ec able]), sugges ing ha dinc eased
wi h de elopmen . dwas s able om baseline o ollow-up,
al hough sligh ly less so han IQ was (Wechsle Abb e ia ed
Scale o In elligence, WASI) ( = 0.68, p 0.0 o d; 0.77, p 
0.0 o WASI IQ; 95% CI o he di e ence = 0.135 o 0.044;
Figu e 1B). These es ima es gi e a conse a i e es ima e o
disc iminan alidi y o d e sus IQ D = 0.76, which is sa is ac o y
(<0.85) (Voo hees e al., 2016). dinc eased wi h es ing wa e (e -
ec size = 0.38, p 0.0). We ound no e idence he e, o in sub-
sequen analyses, o mo e complex models o age (cu ilinea
e ec s o in e ac ions wi h sex).
We con i med ha bo h ma ix and ocabula y aw IQ sub-
sco es obus ly co ela ed wi h d( ixed-e ec be as = 0.088,
0.179, SE = 0.008, 0.018, p 0.0). Howe e , inclusion o aw
IQ sco es did no a ec he signi icance o age as a eg esso
(age be a = 0.121, SE = 0.020, p 0.0). The e o e, no only did
decision acui y inc ease wi h age in ou sample bu so did he
componen ha was independen o IQ abili ies, sugges ing
ha IQ and dde eloped in pa allel wi h age. Toge he , IQ sub-
sco es and age accoun ed o
2adj
= 0.31 o he a iance in
da baseline.
Wi h espec o sel - epo ed sex, dsco es o males we e
highe han hose o emales a baseline ( es p = 8.6e–5, e -
ec size = 0.27). Howe e , i bo h IQ subsco es and age we e
en e ed in LME, he co ela ion be ween dand sel - epo ed
sex was no longe signi ican . Thus, any unco ec ed sex
dependence is likely o be due o pa icipan sel -selec ion.
Tha is, among males, mo e pa icipan s o highe IQ olun-
ee ed ela i e o among emales. dshowed no signi ican
Figu e 1. Decision acui y
(A) Decision acui y common ac o o e cogni i e
pa ame e s, based on he alida ed ou - ac o so-
lu ion. Measu e labels a e sho ened e sions o
desc ip ions in Table 1, and le e s in b acke s a e
ask labels e e ing o Table 1. The op hal o a -
iables load posi i ely, while g ay e ical lines gi e a
isual indica ion o which measu es a e impo an ,
being he h esholds used o inclusion o a iables
in he con i ma o y analyses.
(B) Decision acui y was s ongly co ela ed be ween
baseline and ollow-up, as expec ed o a disposi-
ional measu e. Mau e is he eg ession line, and
black is he iden i y line.
age 3sex dependence (con olling o
IQ, sex p = 0.39, age 3sex p = 0.21).
As o socioeconomic ac o s a ec ing
he de elopmen o d, we no ed an in-
c ease wi h pa en al educa ion (p =
0.0051, be a = 0.19, SE = 0.067) bu no signi ican associa ion
wi h neighbo hood dep i a ion (p = 0.09).
Men al heal h ac o s and hei associa ion wi h
decision acui y
Nex , we examined he ela ionship be ween dand bo h psycho-
logical disposi ions and symp oms. No e ha in ou s udy,
in ol ing mainly heal hy adolescen s and young adul s, symp om-
a ology e e s o he na u e and ex en o sel - epo ed men al
heal h symp oms a he han diagnosable clinical diso de s.
Thus, we used ac o sco es alida ed speci ically o ou sample
(Polek e al., 2018;S Clai e al., 2017), which indica ed ha dis-
posi ions and symp oms in ou sample we e well desc ibed by bi-
ac o models. Each bi ac o model comp ises a supe o dina e
‘‘gene al ac o ’’ and subo dina e ‘‘speci ic ac o s.’’ Disposi ions
comp ise a gene al social unc ioning ac o (‘‘sociali y’’) and ou
speci ic ac o s: social sensi i i y, sensa ion seeking, e o ul con-
ol, and suspiciousness. Symp oms comp ise a gene al dis ess
ac o , a.k.a. ‘‘p ac o ’’ (Caspi e al., 2014;Pa alay e al., 2015),
and i e speci ic ac o s: mood, sel -con idence, wo y, abe an
hinking, and an isocial beha io .
dwas signi ican ly p edic ed by disposi ions, o e and abo e
i s ela ionship wi h in elligence. We i s eg essed all symp om
disposi ion ac o sco es agains d, allowing all ac o s o
compe e in explaining a iance in LME models wi h pa icipan
in e cep as andom e ec . dwas signi ican ly and posi i ely
ela ed o he gene al disposi ion ac o , sociali y (p = 0.0002,
s anda dized be a, a.k.a. bz = 0.36, SE(bz) = 0.096). In models
ha included aw IQ sco es and age, bo h a iables signi ican ly
p edic ed dand imp o ed model i (Baysian In o ma ion C i e-
ion, a.k.a BIC = 4,873 e sus 5,083 wi hou IQ). Impo an ly, in-
clusion o IQ s eng hened he signi icance o sociali y (p =
0.0001, bz = 0.32, SE(bz) = 0.084; see Table 2).
Among symp om sco es, dwas mos s ongly associa ed wi h
abe an hinking, which d aws on schizo ypy and obsessional-
i y. Co a ying o IQ, bu no disposi ions, showed ha dsigni i-
can ly dec eased wi h highe abe an hinking (p = 0.016,
be a = 0.16, SE = 0.066), highe gene al dis ess (p = 0.048,
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040, June 16, 2021 2029
be a = 0.12, SE = 0.057), bu lowe wo y (p = 0.014, be a =
0.16, SE = 0.063). Howe e , co a ying o sociali y (wi h o
wi hou o he disposi ions) educed he signi icance o abe an
hinking, o end le el (p = 0.074, be a = 0.10, SE = 0.053),
and abolished he ela ionship wi h o he symp om dimensions
(symp om gene al ac o : dis ess, p = 0.82, o he s anging
om p = 0.35 o 0.99). By i sel , IQ was signi ican ly co ela ed
o abe an hinking (ma ix p = 0.013, ocabula y p = 0.0001)
and less so gene al dis ess (ma ix p = 0.012, ocabula y p =
0.47). Again, all analyses linea ly accoun ed o age and did no
bene i om mo e complex models o age.
Pa e ns o b ain connec i i y a e associa ed wi h
decision acui y di e en ly om IQ
Ou o 313 heal hy subjec s who we e scanned a baseline, we
disca ded baseline scans wi hou accep able imaging da a qual-
i y (3), whose ME-ICA denoising did no con e ge (4), o who had
excessi e mo ion while scanning (8), lea ing 298 baseline scans
o analysis. A u he h ee subjec s we e emo ed om ana-
lyses in ol ing IQ sco es as hey did no comple e he IQ es s,
lea ing 295 subjec s o analysis. A popula ion-a e age pa cella-
ion o b ain da a was ob ained using independen componen
analysis in ou sample, esul ing in 168 ne wo ks (nodes) wi hin
each o which ac i i y was highly co ela ed. Pa e ns o connec-
i i y be ween nodes we e hen es ima ed as pa ial co ela ion
alues, o es ing-s a e unc ional connec i i y ( sFC). We hen
used sFC alues as ea u es in spa se pa ial leas -squa es
(SPLS) analyses o p edic decision acui y and composi e IQ.
We used c oss- alida ion and ou -o -sample p edic i e es ing
o p e en o e i ing. P edic i e accu acy was assessed as
Pea son’s co ela ion coe icien be ween ue sco es and
model-p edic ed alues. We epo associa ions be ween
p edic ed and obse ed decision acui y a e co ec ing o
scanne - ela ed and o he co a ia es. This ensu es ha i is he
in o ma ion ca ied by he unc ional connec i i y alone ha p e-
dic s cogni i e abili ies. (See STAR Me hods o de ails; Figu e 2
illus a es he s uc u e o he p edic i e es ing.)
Sco es o dp edic ed on he basis o unc ional connec i i y,
d
p
, signi ican ly co ela ed wi h measu ed dcon olling o de-
mog aphic and imaging- ela ed co a ia es (see STAR Me hods
o de ails; = 0.145, p < 10
6
). The co ela ion be ween
measu ed IQ and IQ p edic ed on he basis o sFC using all con-
nec ions was lowe bu also signi ican ( = 0.092, p = 9e–5).
To in e p e he neu oana omical s uc u e o he p edic i e
model, we i s pa i ioned he nodes in o ana omically meaning-
ul ‘‘modules’’ using a communi y de ec ion algo i hm (Blondel
e al., 2008) and hen asked how well each o hese modules p e-
dic ed d. The communi y de ec ion algo i hm clus e ed he no-
des in o disjoin communi ies o modules based on he s eng h
o hei in insic connec i i y, o some ex en analogous o la ge-
scale unc ional ne wo ks. As shown in Figu e 2, we ob ained he
ollowing modules: an e io empo al co ex including he medial
empo al lobe (ATC); on al pole (FPL); on opa ie al con ol
ne wo k (FPN); le do sola e al p e on al co ex (LDC); medial
p e on al co ex (MPC); o bi o on al co ex, medial and la e al
(OFC); ope cula co ex (OPC); pos e io cingula e co ex
(PCC); pos e io empo al co ex (PTC); igh do sola e al p e-
on al co ex (RDC); subco ical (SUB); salience ne wo k
(SAN); soma osenso y and mo o a eas (SMT); and isual e-
gions (VIS). We i ed a di e en SPLS model o he subse o
Table 2. Key s eps in eg ession analyses
Independen a iable
A. Symp oms only (p alue
o ixed e ec s be a; ime-
dependen LME)
B. Disposi ions only (p alue
o be a; baseline only)
C. Symp oms and disposi ions
(p alue o ixed e ec s be a;
ime-dependen LME)
Gene al symp om ac o :
Gene al dis ess
0.048*– 0.390
Sel -con idence speci ic
ac o (SF)
0.351 – 0.316
An isocial beha io SF 0.381 – 0.912
Wo y SF 0.014*– 0.875
Abe an hinking SF 0.016*– 0.074
#
Mood SF 0.813 – 0.871
Gene al disposi ion ac o :
Adap i e sociali y
– 0.0018** 0.0001***
Social sensi i i y – 0.656 –
Sensa ion seeking – 0.987 –
E o ul con ol – 0.959 –
Suspiciousness – 0.014*–
Age <0.0001*** 0.0002*** <0.0001***
Vocabula y IQ ( aw sco e) <0.0001*** <0.0001*** <0.0001***
Ma ix IQ ( aw sco e) <0.0001*** <0.0001*** <0.0001***
*signi ican a p = 0.05.
**signi ican a p = 0.005.
***signi ican a p < 0.001.
#
end le el signi icance a p = 0.05.
ll
OPEN ACCESS A icle
2030 Neu on 109, 2025–2040, June 16, 2021
connec ions in ol ing he nodes in each module, including bo h
in a- and in e modula connec ions.
The co ela ion be ween measu ed and p edic ed dsco es
was signi ican o he FPN, MPC, OFC, OPC, PCC, SMT,
and VIS modules a e co ec ion o mul iple es s (Figu e 4A;
Table 2), wi h he s onges co ela ions o OFC, PCC, and
SMT. Fo he PCC and SMT modules, he co ela ion coe i-
cien s exceeded o a small deg ee he co ela ion o a model
employing all possible connec ions. This can bes be explained
as a esul o ea u e selec ion. In he ull model, i is ha de o
selec jus he igh ea u es and p o ec agains o e - i ing, e-
sul ing in a g ea e penal y in p edic i e accu acy. On he o he
hand, he model ained on a smalle se o ea u es alone is
less likely o o e i . This pa adoxical inc ease in accu acy o
a model wi h less ea u es is known o be s onge when he
numbe o obse a ions is small, ela i e o he numbe o ea-
u es (Chu e al., 2012), which is he case in ou da ase . The
di e en modules comp ised di e se numbe s o nodes, bu
he e was no signi ican associa ion be ween he numbe o
model ea u es and he co ela ion be ween obse ed and p e-
dic ed sco es (d: = 0.356, p = 0.193; IQ composi e sco es: =
0.158, p = 0.574).
Ou o 235 subjec s who we e scanned a ollow-up, adhe ing
o he same c i e ia as o he baseline da a, we disca ded hose
wi hou accep able imaging da a quali y (4), whose ME-ICA
denoising did no con e ge (5), and who p esen ed wi h exces-
si e mo ion (3), lea ing 223 subjec s a ailable o analysis. We
applied he model ained on he baseline da a o he ollow-up
da a (see STAR Me hods) o he modules whe e he p edic ion
was signi ican a baseline. Impo an ly, he p edic ion o a sub-
jec a ollow-up did no in ol e hei own sFC baseline da a, as
his would in la e he es ima e o p edic i e pe o mance. The
baseline model p edic ed signi ican ly he ollow-up d alues
based on he ollow-up connec i i y da a when using ei he all
he connec ions o hose wi h ne wo ks in he FPN, MPC,
OFC, and SMT modules, con olling o demog aphic and imag-
ing- ela ed co a ia es, and co ec ing o mul iple es s (Fig-
u e 4B; Table 3).
To assess whe he dand IQ can be p edic ed by speci ic
sFC pa e ns o , al e na i ely, whe he bo h a e unde pinned
Figu e 2. S uc u e o p edic i e es ing
Flow diag am o he nes ed c oss- alida ion pipeline used o es ima e how s ongly decision acui y (simila ly o IQ) could be p edic ed om b ain da a.
Essen ially, a p edic i e model was de i ed om aining olds and hen applied o he b ain da a om es olds o de i e p edic ed alues o he decision acui y
o each indi idual. This could hen be compa ed wi h he expe imen ally de i ed decision acui y. In ou s udy, N
B
= 200, N
F1
=20,N
F2
= 10, N
R
= 5, and N
P
= 100. X
co esponds o he sFC ea u es and y o he sco es p edic ed (do IQ).
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040, June 16, 2021 2031
by simila pa e ns o neu al connec i i y, we con olled he
pa ial co ela ion coe icien s be ween d
p
and d,on opo
he nuisance co a ia es p e iously included, o IQ. In a com-
plemen a y manne , we con olled he pa ial co ela ion be-
ween IQ
p
and IQ, on op o he nuisance co a ia es, o d.A -
e co ec ion o IQ composi e sco es, and co ec ing o
mul iple compa isons, he co ela ion be ween dand d
p
e-
mained signi ican o OPC, PCC, and SMT (Figu e 5A; Table
2), sugges ing ha hese modules e lec decision acui y
o e and abo e hei ela ion o IQ. On he o he hand, he co -
ela ion be ween IQ
p
and IQ was signi ican o OPC and PTC
a e con olling o d(Figu e 5B; Table 3), sugges ing ha
hese modules e lec IQ o e and abo e hei ela ion o de-
cision acui y. These analyses demons a e ha decision acu-
i y and IQ ha e dis inguishable and speci ic signa u es in
unc ional connec i i y ne wo ks: decision acui y aps on he
de aul mode, salience, and senso imo o ne wo ks, whe eas
IQ aps on he salience ne wo k bu also on empo al ne -
wo ks associa ed wi h language p ocessing.
DISCUSSION
To ou knowledge, his is he i s s udy cha ac e izing a dimen-
sional s uc u e in co e decision-making om an epidemiologi-
cally in o med sample o adolescen s and young adul s. We
ound ha decision-making pe o mance could be desc ibed
by a b oad cons uc ecei ing con ibu ions om mul iple do-
mains o cogni ion. We e med his decision acui y, d. In ou
sample, dshowed sa is ac o y longi udinal s abili y, inc eased
wi h age and wi h IQ. dalso had speci ic associa ions wi h
men al heal h measu es, o e and abo e IQ. Decision acui y
was ela ed o b ain unc ion, showing a empo ally s able asso-
cia ion wi h sFC, in ol ing ne wo ks p e iously implica ed in de-
cision-making p ocesses. Mo eo e , sFC pa e ns associa ed
wi h dand IQ we e dis inguishable and speci ic despi e showing
a deg ee o o e lap.
Decision acui y had an in e p e able s uc u e, e lec ing a a-
cili y o good decision-making. Decision acui y inc eased as
decision a iabili y lessened, e idenced by i s loadings on
Figu e 3. B ain ne wo ks
Modules de ec ed by he communi y s uc u e algo i hm. The 168 nodes o he pa cella ion we e clus e ed in 14 modules wi h high a e age sFC among
hei nodes. ATC, an e io empo al co ex including he medial empo al lobe; FPL, on al pole; FPN, on opa ie al con ol ne wo k; LDC, le do sola e al
p e on al co ex; MPC, medial p e on al co ex; OFC, o bi o on al co ex, medial and la e al; OPC, ope cula co ex; PCC, pos e io cingula e co ex;
PTC, pos e io empo al co ex; RDC, igh do sola e al p e on al co ex; SUB, subco ical; SAN, salience ne wo k; SMT, soma osenso y and mo o
a eas; VIS, isual egions.
ll
OPEN ACCESS A icle
2032 Neu on 109, 2025–2040, June 16, 2021
Pa alay, P., Fonagy, P., Deigh on, J., Belsky, J., Vos anis, P., and Wolpe , M.
(2015). A gene al psychopa hology ac o in ea ly adolescence. B . J.
Psychia y 207, 15–22.
Paus, T., Kesha an, M., and Giedd, J.N. (2008). Why do many psychia ic dis-
o de s eme ge du ing adolescence? Na . Re . Neu osci. 9, 947–957.
Payzan-LeNes ou , E., Dunne, S., Bossae s, P., and O’Dohe y, J.P. (2013).
The neu al ep esen a ion o unexpec ed unce ain y du ing alue-based deci-
sion making. Neu on 79, 191–201.
Pea son, J.M., Heilb onne , S.R., Ba ack, D.L., Hayden, B.Y., and Pla , M.L.
(2011). Pos e io cingula e co ex: adap ing beha io o a changing wo ld.
T ends Cogn. Sci. 15, 143–151.
Pea son, R.M., He on, J., Bu on, K., Ben all, R.P., Fe nyhough, C., Mahedy,
L., Bowes, L., and Lewis, G. (2015). Cogni i e s yles and u u e dep essed
mood in ea ly adul hood: The impo ance o global a ibu ions. J. A ec .
Diso d. 171, 60–67.
Phelps, E.A., Lempe , K.M., and Sokol-Hessne , P. (2014). Emo ion and deci-
sion making: mul iple modula o y neu al ci cui s. Ann. Re . Neu osci. 37,
263–287.
Polek, E., Jones, P.B., Fea on, P., B odbeck, J., Mou oussis, M., Dolan, R.,
Fonagy, P., Bullmo e, E.T., and Goodye , I.M.; Nspn Conso ium (2018).
Pe sonali y dimensions eme ging du ing adolescence and young adul hood
a e unde pinned by a single la en ai indexing impai men in social unc-
ioning. BMC Psychia y 18,23.
Powe , J.D., Cohen, A.L., Nelson, S.M., Wig, G.S., Ba nes, K.A., Chu ch, J.A.,
Vogel, A.C., Laumann, T.O., Miezin, F.M., Schlagga , B.L., and Pe e sen, S.E.
(2011). Func ional ne wo k o ganiza ion o he human b ain. Neu on 72,
665–678.
Powe , J.D., Ba nes, K.A., Snyde , A.Z., Schlagga , B.L., and Pe e sen, S.E.
(2012). Spu ious bu sys ema ic co ela ions in unc ional connec i i y MRI ne -
wo ks a ise om subjec mo ion. Neu oimage 59, 2142–2154.
Rigoli, F., Ru ledge, R.B., Chew, B., Ousdal, O.T., Dayan, P., and Dolan, R.J.
(2016). Dopamine Inc eases a Value-Independen Gambling P opensi y.
Neu opsychopha macology 41, 2658–2667.
Rosenbe g, M.D., Finn, E.S., Scheinos , D., Papademe is, X., Shen, X.,
Cons able, R.T., and Chun, M.M. (2016). A neu oma ke o sus ained a en ion
om whole-b ain unc ional connec i i y. Na . Neu osci. 19, 165–171.
Rubino , M., and Spo ns, O. (2010). Complex ne wo k measu es o b ain con-
nec i i y: uses and in e p e a ions. Neu oimage 52, 1059–1069.
Rushwo h, M.F.S., Noonan, M.P., Boo man, E.D., Wal on, M.E., and Beh ens,
T.E. (2011). F on al co ex and ewa d-guided lea ning and decision-making.
Neu on 70, 1054–1069.
Scholl, J., and Klein-Fl€
ugge, M. (2018). Unde s anding psychia ic diso de by
cap u ing ecologically ele an ea u es o lea ning and decision-making.
Beha . B ain Res. 355, 56–75.
Schuck, N.W., Cai, M.B., Wilson, R.C., and Ni , Y. (2016). Human
O bi o on al Co ex Rep esen s a Cogni i e Map o S a e Space.
Neu on 91, 1402–1412.
Shaha , N., Hause , T.U., Mou oussis, M., Mo an, R., Ke ama i, M., and
Dolan, R.J.; NSPN conso ium (2019a). Imp o ing he eliabili y o
model-based decision-making es ima es in he wo-s age decision
ask wi h eac ion- imes and d i -di usion modeling. PLoS Compu .
Biol. 15, e1006803.
Shaha , N., Mo an, R., Hause , T.U., Kie i , R.A., McNamee, D.,
Mou oussis, M., and Dolan, R.J.; NSPN Conso ium (2019b). C edi
assignmen o s a e-independen ask ep esen a ions and i s ela ion-
ship wi h model-based decision making. P oc. Na l. Acad. Sci. USA
116, 15871–15876.
Smi h, S.M., Jenkinson, M., Wool ich, M.W., Beckmann, C.F., Beh ens,
T.E.J., Johansen-Be g, H., Bannis e , P.R., De Luca, M., D obnjak, I.,
Fli ney, D.E., e al. (2004). Ad ances in unc ional and s uc u al MR im-
age analysis and implemen a ion as FSL. Neu oimage 23 (Suppl. 1 ),
208–219.
Smi h, S.M., Fox, P.T., Mille , K.L., Glahn, D.C., Fox, P.M., Mackay, C.E.,
Filippini, N., Wa kins, K.E., To o, R., Lai d, A.R., and Beckmann, C.F. (2009).
Co espondence o he b ain’s unc ional a chi ec u e du ing ac i a ion and
es . P oc. Na l. Acad. Sci. USA 106, 13040–13045.
Smi h, S.M., Mille , K.L., Salimi-Kho shidi, G., Webs e , M., Beckmann, C.F.,
Nichols, T.E., Ramsey, J.D., and Wool ich, M.W. (2011). Ne wo k modelling
me hods o FMRI. Neu oimage 54, 875–891.
Smi h, S.M., Nichols, T.E., Vidau e, D., Winkle , A.M., Beh ens, T.E.J.,
Glasse , M.F., Ugu bil, K., Ba ch, D.M., Van Essen, D.C., and Mille , K.L.
(2015). A posi i e-nega i e mode o popula ion co a ia ion links b ain connec-
i i y, demog aphics and beha io . Na . Neu osci. 18, 1565–1567. h ps://doi.
o g/10.1038/nn.4125.
S ipada, C.S., Angs ad , M., Banks, S., Na han, P.J., Libe zon, I., and Phan,
K.L. (2009). Func ional neu oimaging o men alizing du ing he us game in
social anxie y diso de . Neu o epo 20, 984–989.
S ipada, C., Angs ad , M., Ru he o d, S., Taxali, A., G ea house, T., Cla k, D.A.,
Hyde, L., Weiga d, A., B islin, S., Hicks, B., and Hei zeg, M. (2020). Connec omic
Al e a ions Linked o T ansdiagnos ic Risk o Psychopa hology a he T ansi ion
o Adolescence. bioRxi . h ps://doi.o g/10.1101/2020.08.21.260927.
S Clai , M.C., Neu eld, S., Jones, P.B., Fonagy, P., Bullmo e, E.T., Dolan, R.J.,
Mou oussis, M., Toseeb, U., and Goodye , I.M. (2017). Cha ac e ising he
la en s uc u e and o ganisa ion o sel - epo ed hough s, eelings and be-
ha iou s in adolescen s and young adul s. PLoS ONE 12, e0175381.
S alnake , T.A., Liu, T.-L., Takahashi, Y.K., and Schoenbaum, G. (2018).
O bi o on al neu ons signal ewa d p edic ions, no ewa d p edic ion e o s.
Neu obiol. Lea n. Mem. 153, 137–143.
Sun, Y., Danila, B., Josi
c, K., and Bassle , K.E. (2009). Imp o ed communi y
s uc u e de ec ion using a modi ied ine- uning s a egy. EPL 86, 28004.
h ps://doi.o g/10.1209/0295-5075/86/28004.
Su on, R.S., and Ba o, A.G. (1998). Rein o cemen Lea ning: An In oduc ion
(MIT P ess).
Swa , J.C., F ank, M.J., M€
a€
a €
a, J.I., Jensen, O., Cools, R., and den Ouden,
H.E.M. (2018). F on al ne wo k dynamics e lec neu ocompu a ional mecha-
nisms o educing maladap i e biases in mo i a ed ac ion. PLoS Biol. 16,
e2005979.
Symmonds, M., W igh , N.D., Bach, D.R., and Dolan, R.J. (2011).
Decons uc ing isk: sepa able encoding o a iance and skewness in he
b ain. Neu oimage 58, 1139–1149.
Todd, N., Moelle , S., Aue bach, E.J., Yacoub, E., Flandin, G., and Weiskop ,
N. (2016). E alua ion o 2D mul iband EPI imaging o high- esolu ion, whole-
b ain, ask-based MRI s udies a 3T: Sensi i i y and slice leakage a i ac s.
Neu oimage 124 (P A), 32–42.
Uddin, L.Q. (2015). Salience p ocessing and insula co ical unc ion and
dys unc ion. Na . Re . Neu osci. 16, 55–61.
an de Maas, H.L.J., Dolan, C.V., G asman, R.P.P.P., Wiche s, J.M.,
Huizenga, H.M., and Raijmake s, M.E.J. (2006). A dynamical model o gene al
in elligence: he posi i e mani old o in elligence by mu ualism. Psychol. Re .
113, 842–861.
Van Essen, D.C., Smi h, S.M., Ba ch, D.M., Beh ens, T.E.J., Yacoub, E., and
Ugu bil, K.; WU-Minn HCP Conso ium (2013). The WU-Minn Human
Connec ome P ojec : an o e iew. Neu oimage 80, 62–79.
Vidau e, D., Smi h, S.M., and Wool ich, M.W. (2017). B ain ne wo k dynamics
a e hie a chically o ganized in ime. P oc. Na l. Acad. Sci. USA 114,
12827–12832.
Voo hees, C.M., B ady, M.K., Calan one, R., and Rami ez, E. (2016).
Disc iminan alidi y es ing in ma ke ing: an analysis, causes o conce n,
and p oposed emedies. J. Acad. Ma k. Sci. 44, 119–134.
Wal on, M.E., Beh ens, T.E.J., Buckley, M.J., Rudebeck, P.H., and Rushwo h,
M.F.S. (2010). Sepa able lea ning sys ems in he macaque b ain and he ole o
o bi o on al co ex in con ingen lea ning. Neu on 65, 927–939.
Weiskop , N., Suckling, J., Williams, G., Co eia, M.M., Inks e , B., Tai , R., Ooi,
C., Bullmo e, E.T., and Lu i, A. (2013). Quan i a i e mul i-pa ame e mapping o
R1, PD(*), MT, and R2(*) a 3T: a mul i-cen e alida ion. F on . Neu osci. 7,95.
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040, June 16, 2021 2039

Whi ake , K.J., Ve
´ es, P.E., Rome o-Ga cia, R., Va
´
sa, F., Mou oussis, M.,
P abhu, G., Weiskop , N., Callaghan, M.F., Wags yl, K., Ri man, T., e al.;
NSPN Conso ium (2016). Adolescence is associa ed wi h genomically
pa e ned consolida ion o he hubs o he human b ain connec ome. P oc.
Na l. Acad. Sci. U.S.A. 113, 9105–9110.
Wilson, R.C., Takahashi, Y.K., Schoenbaum, G., and Ni , Y. (2014).
O bi o on al co ex as a cogni i e map o ask space. Neu on 81, 267–279.
Winkle , A.M., Ridgway, G.R., Douaud, G., Nichols, T.E., and Smi h, S.M.
(2016). Fas e pe mu a ion in e ence in b ain imaging. Neu oimage 141,
502–516.
ll
OPEN ACCESS A icle
2040 Neu on 109, 2025–2040, June 16, 2021
STAR+METHODS
KEY RESOURCES TABLE
RESOURCE AVAILABILITY
Lead con ac
Fu he in o ma ion and eques s o esou ces should be di ec ed o and will be ul illed by he Lead Con ac , Michael Mou oussis (m.
[email p o ec ed]).
Ma e ials a ailabili y
No applicable
Da a and code a ailabili y
Due o he wo ding o he consen ha pa icipan s ga e o he NSPN p ojec , all pseudo-anonymized da a suppo ing his s udy is
a ailable upon legi ima e-in e es eques om [email p o ec ed].
REAGENT o RESOURCE SOURCE IDENTIFIER
Deposi ed da a
P ocessed connec i i y
ma ices
This pape h ps://gi hub.com/
benjaminga zon/FCPC/ ee/
mas e /da a
ICA maps and unc ional
modules
This pape h ps://gi hub.com/
benjaminga zon/FCPC/ ee/
mas e /da a
Da a o cogni i e ask ac o
analyses
This pape h ps://gi hub.com/
mmou ou/decAc
ile AllD18.R
Da a o Decision Acui y
londi udinal analyses
This pape h ps://gi hub.com/
mmou ou/decAc
ile sym acdeciq.cs
Sc ip s o cogni i e ask
ac o analyses and Decision
Acui y longi udinal analyze
This pape h ps://gi hub.com/
mmou ou/decAc
R iles CFA-decAc.R and
decAclongi.R ; op ional
u ili ies ile gen_u .R
So wa e and algo i hms
MATLAB Ma hwo ks RRID: SCR_001622; h ps://
www.ma hwo ks.com/
R package The R Founda ion RRID: SCR_001905; h ps://
www. -p ojec .o g
ME-ICA Kundu e al., 2017 h ps://a ni.nimh.nih.go /
pub/dis /s c/pkundu/
README.meica
FSL Smi h e al., 2004 RRID: SCR_002823; h ps://
sl. m ib.ox.ac.uk/ sl/ slwiki/
SPLS R lib a y Chun and Keles¸, 2010 h ps://c an. -p ojec .o g/
web/packages/spls
B ain Connec i i y Toolbox Rubino and Spo ns, 2010 RRID: SCR_004841; h p://
www.b ain-connec i i y-
oolbox.ne
Func ional connec i i y
analysis sc ip s
This pape h ps://gi hub.com/
benjaminga zon/FCPC
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040.e1–e7, June 16, 2021 e1
All code pe aining o he analysis o compu a ional ask measu es and hei ac o analysis is a ailable upon eques om m.
[email p o ec ed], while code pe aining o he unc ional connec i i y analysis is a ailable upon eques om benjamin.
[email p o ec ed].
EXPERIMENTAL MODEL AND SUBJECTS DETAILS
Human subjec s
Pa icipan s we e sampled om a pool o c. 2400 communi y-dwelling young people and o med a ‘cogni i e coho ’. Pa icipan s
we e con ac ed a andom om 5 age bins (14-16,16-18 e c.), un il each ec ui ed age bin had app oxima ely equal p opo ions
o emales and males. The p opo ion o non-whi e-English youngs e s in ou s udy was wi hin 10% o ha o he mos ecen census.
Signi ican neu opsychia ic p oblems we e sc eened ou by sel - epo , and ec ui men sou ces we e selec ed o he sample o be
as ep esen a i e as possible o he heal hy popula ion (Kiddle e al., 2018). We con inued o in i e people om he la ge pool in o he
cogni i e coho , un il ou a ge numbe o 780 ‘cogni i e’ pa icipan s was comple ed. O hese, 300 we e in i ed o MRI b ain scan-
ning. They we e equally dis ibu ed in he 5 age bins as abo e, and equal Female:Male a io. A he ime o egis e ing wi h he s udy,
pa icipan s we e asked o ick: Sex: ‘Female’ o ‘Male’. All pa icipan s ha ga e da a o decision acui y and imaging analyses
icked one o he o he box. I was no cla i ied i some unde s ood he ques ion as ‘gende iden i y’, socially a ibu ed o biological
ca ego y. Due o he ph asing ‘Sex:’ we expec ha mos pa icipan s unde s ood he ques ion o mean ‘sel - epo ed es ima e o
biological sex’, bu his is a en a i e in e p e a ion.
In addi ion, hey we e sc eened o absence o a his o y o p esence o men al heal h diso de , neu ological o majo heal h p ob-
lem, o lea ning disabili y. Ini ial sc eening was by sel - epo bu was con i med by SCID-II in e iew and IQ es ing.
We supplemen ed his non-heal hca e-seeking sample wi h 50 young people ecen ly diagnosed wi h DSM-5 majo dep essi e
diso de . O hese, 38 ga e decision-making ba e y da a o decision-acui y analyses (M = 11,F = 27). Thus, he main sample
was ep esen a i e o he heal hy wide popula ion, bu a smalle dep ession g oup was also analyzed o es whe he he s uc u e
o decision-making and he ele an b ain measu es iden i ied in he heal hy popula ion also ex ended o his heal h-seeking g oup.
The dep essed coho was excluded om MRI analyses epo ed he e.
Pa icipan s (and hei pa en s, i less han 16 yea s old) ga e in o med consen o pa icipa e in he s udy. The s udy was app o ed
by he Camb idge E hics Commi ee (12/EE/0250).
METHOD DETAILS
Sample size es ima ion
Ou key sample size es ima ion pe ained o he neu oimaging sample, and esul ed in he es ima e o N = 300. The cogni i e- ask
sample was hen as la ge as s udy esou ces allowed, including esou ces needed o e- elephone pa icipan s who had ini ially gi en
consen bu did no immedia ely espond o an in i a ion o ollow-up, up o achie ing a ollow-up a e o a leas 70%. In summa y,
es ima ion o he key, neu oimaging coho sample size p oceeded as ollows.
A he ime o s udy design, he e we e no speci ic s udies o p o ide a igo ous analysis o sFC de elopmen al, longi udinal sam-
ple size es ima ion. We he e o e elied on a oughly compa able s udy which allowed o imaging de elopmen al e ec . This s udy
used a coho o 387 pa icipan s, who p o ided 829 s uc u al MRI scans (Giedd, 2004). We hus aimed o 300 pa icipan s, a num-
be which was logis ically accessible, and op imized powe by selec ing pa ame e s (age minimum and wid h) o age-bins and ollow-
up in e als, using published g ay-ma e olume da a as a p oxy o he indi idual a ia ion ha we should ha e powe o de ec .
Quad a ic g ow h cu es we e i ed o he da a om he published s udy abo e, and s udy pa ame e s a ied in silico o minimize
a iance o he es ima ed pa ame e s o he g ow h-cu es. Simula ions showed a pla eauing o e iciency i he o e all age ange was
educed o less han 10 yea s, o he wid h o age-bins o less han 2 yea s. Pa ame e accu acy imp o ed wi h ollow-up in e al and
de e io a ed i he ollow-up was sho e han 6 mon hs. The e o e, we aimed o 5 age bins imes 2 yea s wid h, and selec ed a min-
imum in e al o 12 mon hs, aiming a abou 18 mon h a e age. This was well abo e 6 mon hs, educed he chance o demog aphic
loss (mo ing a away) and allowed adequa e ime o epea in i a ions o pa icipan s ha did no immedia ely espond o ollow-up
in i a ions.
Decision-making Task Ba e y
We selec ed se en asks apping undamen al decision-making wi h e idence linking hem o bo h men al heal h symp oms and neu-
al mechanisms (Table 1 in main ex ). Fi s , a Go-NoGo ask (Gui a -Masip e al., 2012) p o ided measu es ele an o sensi i i y o
ewa ds and Pa lo ian bias. Second, an app oach-a oidance ask measu ed he balance o seeking ewa ds e sus a oiding losses
(Bach e al., 2014). This is likely o be ele an o e e yday isk- aking by young people. Thi d, a isk p e e ence ask (Symmonds e al.,
2011) complemen ed his, ocusing on widely accep ed economic measu es o isk- aking (Bach e al., 2020;Rigoli e al., 2016).
Fou h, we assessed in e - empo al discoun ing, lea ning abou he p e e ences o o he s and inally pee in luence (Mou oussis
e al., 2016;Nicolle e al., 2012). Discoun ing has been shown o be impo an in a ange o psychia ic diso de s (Bickel e al.,
2012) and so a e issues o hinking abou o he s (S ipada e al., 2009) and pee in luence (Ke e al., 2012). Fi h, we included an in-
o ma ion ga he ing ask (Mou oussis e al., 2011) as his has been consis en ly shown o be ele an o psycho ic symp oms (Lincoln
ll
OPEN ACCESS A icle
e2 Neu on 109, 2025–2040.e1–e7, June 16, 2021
e al., 2010b) as well as he undamen als o decision-making (Dayan, 2014). Six h, a T us Task was used as a measu e o complex
social cogni ion especially ele an o diso de s o in e pe sonal unc ion (Fe e al., 2012;King-Casas e al., 2008). Se en h, a wo
s ep ask assessed he ole o habi ual e sus plan ul mechanisms in decision-making (Daw e al., 2011). The ba e y was imple-
men ed using MATLAB (MATLAB, 2012) using he Cogen oolbox (see acknowledgmen s). T ained esea ch assis an s di ec ed
he pa icipan s h ough he ba e y.
In e ms o emune a ion, pa icipan s ecei ed a la ee bu we e also ( u h ully) old ha hey would be paid ex a acco ding o
hei ea nings in he asks. They we e in o med ha he e would be a subs an ial amoun o luck in each ask, bu hose who
comple ed he asks ca e ully would expec o ea n abou 2.5 pounds ex a pe ask. Pa icipan s did no see ea nings o each ial,
because asks di e ed g ea ly in hei deli e y and we did no wan o display a ying amoun s o money o a oid addi ional Pa lo ian
mo i a ional e ec s. Ins ead, pa icipan s we e old ha ‘ oughly, each good decision in each ask is wo h app oxima ely he same’,
a s a emen which p o ided a easonable e lec ion o he ue s a e o a ai s. The sole elemen o decep ion in he ba e y was ha
du ing he in e pe sonal asks pa icipan s we e old ha hei play pa ne was a pee , whe eas in eali y i was a compu e agen .
Howe e hese agen s we e simula ing as closely as possible he pe o mance o heal hy people who had he same demog aphics as
he pa icipan s. Pa icipan s we e deb ie ed a he end o all es ing.
Ea nings we e added o hei compensa ion o he day’s es ing, excep o he In e pe sonal-Discoun ing ask. He e, pa icipan s
we e paid a one o hei chosen delays, andomly chosen om all he ials in he ask, i hey chose a la ge bu delayed paymen .
This was paid in Amazon ouche s.
The o de o he asks was subjec o cons ained andomiza ion. We i s pilo ed he ba e y in 15 pa icipan s, o whom we asked
de ailed eedback as o how in e es ing and how i ing hey ound each ask, as well as ee- o m commen s. On he basis o his we
a oided pu ing he mo e i ing o less in e es ing asks nea he end o he ba e y, in o de o minimize he e ec o a igue. This
esul ed in eigh di e en ask sequences, one o which was gi en a andom o pa icipan s. A e he i s 40 pa icipan s we e e-
c ui ed we pe o med an in e im analysis o compa e pe o mance in his ba e y o sho ened asks as compa ed o he ull-leng h
e sions. Pe o mance in each ask showed ollowed he pa e n o pe o mance in he o iginal, excep he Two-S ep ask. He e pa -
icipan s as a g oup showed only jus -de ec able goal-di ec ed decision-making. As his would g ea ly educe he ask’s use ulness
we imp o ed he p e- ask aining and ins uc ions and disca ded his i s 10% o da a o his ask, wi h sa is ac o y esul s.
Tasks las ed 8-30 min each, gi ing an o e all du a ion o 2 3
=
4–23
=
4h, including one obliga o y b eak and as many ex a be ween-
ask b eaks as he pa icipan asked o . Good pe o mance a ac ed p opo ionally g ea e ees in eal money.
Key measu es we e i s ex ac ed om each ask acco ding o published me hodologies. These key measu es assess unda-
men al aspec s o decision-making, namely sensi i i y o ewa ds and losses, a i udes o isk, in e - empo al and e lec ion impul-
si i y, p o-sociali y and model-basedness. 820 pa icipan s (including all scanned pa icipan s) yielded usable da a ac oss asks. The
app oach-a oidance ask, he in o ma ion ga he ing ask, and he us ask equi ed some adap a ions ha a e lis ed below.
We we e in e es ed in whe he common ac o s ope a ed ac oss domains o decision-making. We he e o e p e-p ocessed he
da a o educe s ong co ela ions among measu es wi hin- ask, which would o he wise domina e he ac o analysis, as is desc ibed
in he supplemen al in o ma ion. In o al we o med 32 measu es, lis ed in Tables 1 and S1.
The app oach a oidance ask was o iginally desc ibed in Bach e al. (2014) was adap ed o he pu poses o his s udy. Because o
ime cons ain s we educed he numbe o h ea con ex s om h ee o wo, which we call wo ‘p eda o s’ co esponding o low and
high h ea . Also, di e en om he p e ious s udy, epoch du a ion did no depend on h ea le el. Tha is, an epoch ended a e a
andom du a ion, independen o whe he he p eda o woke up o no . Finally, he numbe o epochs was educed o 1/3 o he o ig-
inal, so ha he ask ook abou 23 min o comple e.
Based on he p e ious wo k (Bach e al., 2014), we collec ed a la ge numbe o beha io al desc ip i e measu es and pe o med an
explo a o y ac o analysis o hese (subs an ially co ela ed) measu es. We ound ha he i s h ee ac o s could be meaning ully
in e p e ed in decision-making e ms, namely as sensi i i y o he le el o h ea in he en i onmen (‘ h ea sensi i i y’), sensi i i y o
ea u es inc easing p obabili y o loss wi hin an en i onmen (‘loss sensi i i y’) and measu es o o e all pe o mance (‘pe o mance’).
As migh be expec ed, his hi d ‘pe o mance’ ac o loaded mo e highly in d(Figu e 1) bu s ill did no exceed he h eshold o 0.25
ha we used o inclusion in con i ma o y analyses (below).
The ‘co e s o y’ and g aphics o he In o ma ion Ga he ing ask we e adap ed om he wo k o Lincoln and cowo ke s (Lincoln
e al., 2010a,2010b). On he basis o p e ious wo k (Mou oussis e al., 2011) we educed he maximum numbe o samples o
in o ma ion pe ial in o de o inc ease he impac o he app oaching end (u gency). We i s p esen ed pa icipan s wi h an un-
cos ed-in o ma ion ga he ing e sion o he ask o 10 ials. This had delibe a ely non-speci ic ins uc ions o maximize he chance
ha pa icipan s would b ing hei own, subjec i e cos s uc u e o bea and because, somewha unexpec edly, such uncos ed,
sca ce-ins uc ion e sions o he ask has p oduced some o he mos consis en esul s in clinical and subclinical samples. We
hen p esen ed hem wi h 10 ials wi h mo e speci ic ins uc ions. Pa icipan s s a ed wi h 100 poin s and had o pay 10 poin s
o each i em o in o ma ion hey eques ed. We employed a maximum-likelihood i o he bayesian-obse e model om Mou oussis
e al. (2011).
In o de o analyze he T us ask we adap ed he measu es desc ibed by Fe e al. (2012). Following hese esea che s, we consid-
e ed whe he pa icipan s inc eased o dec eased hei o e a each mo e in esponse o obse ing hei pa ne inc ease o
dec ease hei s. Howe e we conside ed he ac ional change in con ibu ion, i.e., he change in he ac ion o play-money ha
could ha e been gi en. This en ails he hypo hesis ha each playe conside s he o he as ‘messaging’ hem om a baseline o hei
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040.e1–e7, June 16, 2021 e3
inancial means, no in absolu e e ms. We hen conside ed he ec o in he 2-dimensional space o ( ac ional-change-o -In es o
by ac ional-change-o -T us ee) o med o each ound o play. We classi ied his in he same way as Fe e al. ( e alia ing, epai ing,
hono ing, dis up ing) as he angle be ween he ec o and he change-o -In es o axis inc eased om 180 o 180 deg ees. Again
using he ( a he c ude) app oxima ion ha s a egy emains he same h oughou he 10 ounds o he game, we added he ec o s
o each o he ounds o de e mine he cha ac e o he game as a whole. The o ien a ion o he esul an ec o cha ac e izes he
whole exchange – bo h In es o (ou pa icipan ) and T us ee ( he compu e ). As all in es o s played he same compu e p og am, his
ec o can be seen as he ype o exchange ha he pa icipan elici ed.
In he e en , o ien a ions showed a clea bimodal dis ibu ion, ei he a ound ze o deg ees (an exchange based on coaxing he
T us ee) o a ound 3p/4. The la e ep esen s an exchange whe e each pa y is esponding o he o he ’s educ ion in con ibu ion
wi h hei own educ ion. We migh specula e ha pa icipan s a emp o signal ‘i you won’ be gene ous, I won’ ei he ’. The wo-
clus e dis ibu ion could in u n be i ed easonably well wi h a single s aigh line spanning e alia o y o coaxing exchanges. The
‘ us building’ index in Tables 1 and 2co esponds o he pa icipan ’s posi ion along his line.
MRI da a acquisi ion
MRI scans we e acqui ed on h ee iden ical 3T whole-body MRI sys ems (Magne om TIM T io; VB17 so wa e e sion; Siemens
Heal hca e): wo loca ed in Camb idge and one loca ed in London. Reliabili y o he MRI p ocedu es ac oss si es has been demon-
s a ed elsewhe e (Weiskop e al., 2013). S uc u al MRI scans we e acqui ed using a mul i-echo acquisi ion p o ocol wi h six equi-
dis an echo imes be ween 2.2 and 14.7 ms, and a e aged o o m a single image o inc eased signal- o-noise a io (SNR); TR =
18.70 ms, 1.0 mm iso opic oxel size, ield o iew (FOV) = 256 3256, and 176 sagi al slices wi h pa allel imaging using GRAPPA
ac o 2 in an e io -pos e io phase-encoding di ec ion. Res ing-s a e blood-oxygen-le el dependen (BOLD) MRI ( s MRI) da a we e
acqui ed using mul i-echo acquisi ion p o ocol wi h h ee echo imes (TE = 13, 31, 48 ms), TR o 2420 ms, 263 olumes, 3.8 mm
iso opic oxel size, 34 oblique slices wi h sequen ial acquisi ion and a 10% gap, FOV = 240 3240 mm and ma ix size = 64 3
64 334. The du a ion o he unc ional scan was app oxima ely 11 min.
Connec i i y Analysis
The s MRI da a we e denoised wi h mul i-echo independen componen analysis (ME-ICA) (Kundu e al., 2017). ME-ICA le e ages
he echo ime dependence o he BOLD signal o sepa a e BOLD- ela ed om a i ac ual signal sou ces, like head mo ion. The unc-
ional images we e no malized o MNI space by composing a igid ans o ma ion o he a e age unc ional image o he pa icipan ’s
s uc u al image and a non-linea ans o ma ion o he s uc u al image o he MNI empla e, and inally smoo hed wi h a 5 mm ull-
wid h-a -hal -maximum Gaussian ke nel. Following Smi h e al. (2015), g oup-ICA was applied o he p e-p ocessed MRI baseline
da a o decompose i in 200 nodes, 32 o which we e iden i ied as a i ac s by isual inspec ion and excluded. The emaining 168
nodes a e ei he con ined b ain egions o ne wo ks o med by egions whe e BOLD signal ime-se ies a e s ongly co ela ed. Mul-
iple spa ial eg essions agains he g oup-ICA spa ial maps we e used o es ima e ime-se ies o each ne wo k and subjec , o bo h
baseline and ollow-up scans. RsFC ma ices (168 3168 nodes) we e hen compu ed using pa ial co ela ion wi h limi ed L2 eg-
ula isa ion (Smi h e al., 2011). All hese p ep ocessing s eps we e conduc ed wi h he ME-ICA oolbox (h ps://a ni.nimh.nih.go /
pub/dis /s c/pkundu/README.meica) and he FMRIB So wa e Lib a y (FSL, h ps:// sl. m ib.ox.ac.uk/ sl/ slwiki/). As shown in Fig-
u e S4 and ecen s udies o FC eliabili y (Noble e al., 2017), o e all eliabili y o indi idual unc ional connec ions is low, hough
some connec ions display mode a e o high eliabili y. We hus used mul i a ia e me hods combining mul iple FC alues as a s a egy
o compensa e o he low FC o indi idual connec ions.
The ob ained sFC alues we e used as ea u es in a spa se pa ial leas -squa es (SPLS) model o p edic wo ou come measu es
o in e es (decision acui y and IQ composi e sco es). SPLS (Chun and Keles¸, 2010; ‘spls’ R lib a y, h ps://c an. -p ojec .o g/web/
packages/spls/) is a mul i a ia e eg ession model ha simul aneously achie es da a educ ion and ea u e selec ion. I has applica-
ion in da ase s wi h highly co ela ed ea u es and sample size much smalle han he o al numbe o ea u es, as was he case in he
p esen s udy. SPLS models a e go e ned by wo pa ame e s (numbe o la en componen s and a h eshold con olling model spa -
si y) ha we e adjus ed using a nes ed c oss- alida ion scheme (i.e., using da a in he aining da ase only) wi h 10- olds (Figu e 2).
P edic ed sco es we e es ima ed by 20- old c oss- alida ion epea ed 5 imes. Fo each aining- es ing pa i ion we pe o med he
ollowing s eps. To elucida e whe he he p edic ions we e d i en by sFC alues independen ly o age, sex o co a ia es o no in-
e es (see below), we i ed a linea model o he aining da ase and eg essed ou om he a ge a iable (in bo h aining and
es ing da ase s) age, sex and hei in e ac ion as well as b ain olume, scanning si e and head-mo ion- ela ed pa ame e s. Head
mo ion is known o o igina e spu ious co ela ions ha bias connec i i y es ima es and he e o e (besides he ME-ICA p ep ocessing
explained abo e) we eg essed ou a e age amewise displacemen (FD), a summa y index o he amoun o in-scanne mo ion (Po-
we e al., 2012), and he deg ees o eedom esul ing om he ME-ICA denoising, which may di e ac oss subjec s depending on
how much nuisance a iance is emo ed om hei da a. As an addi ional con ol o head mo ion, subjec s whose mean FD was
abo e 0.3 mm we e no included in he analysis. We also s anda dized bo h aining and es ing da a wi h espec o he mean
and s anda d de ia ion o he aining da a (sepa a ely o each ea u e). As a i s s ep o il e ou unin o ma i e ea u es and speed
up compu a ions, only hose signi ican ly (p < 0.05) co ela ed wi h he ou come a iable in he aining da ase we e en e ed in he
SPLS model. We hen used a bagging s a egy whe e da a we e esampled wi h eplacemen 200 imes and as many SPLS models
we e i ed o he esampled da ase s, and hei ea u e weigh s a e aged o p oduce a inal model. The pu pose o his s ep was 1) o
ll
OPEN ACCESS A icle
e4 Neu on 109, 2025–2040.e1–e7, June 16, 2021

imp o e he gene alizabili y o he inal a e age model and 2) o allow es ima ion o he s abili y o he ea u e weigh s selec ed. The
inal, a e age model was used o compu e he p edic ed sco es o he es ing pa i ion. The same p ocedu e was epea ed o all
olds o ob ain one p edic ed sco e o each subjec , whe e he p edic ed sco e o each pa icipan depended only on da a om
o he subjec s in he sample. These p ocedu es we e implemen ed wi h R (h ps://www. -p ojec .o g/) and MATLAB (h ps://
www.ma hwo ks.com).
Ne wo k node communi y s uc u e
To enhance ou unde s anding o he ana omical dis ibu ion o he p edic i e connec ions, we pe o med a ‘ i ual lesion’ analysis
(Dubois, e al., 2018), which en ails assessing he pe o mance o he model when i is ained only on subse s o connec ions ins ead
o he ull ensemble. Fi s , we pa i ioned he se o nodes in o disjoin modules o communi ies ( o some ex en analogous o la ge-
scale unc ional ne wo ks; Smi h e al., 2009) o med by nodes which displayed high connec i i y among hem bu lowe connec i i y
wi h nodes in o he modules. We ob ained he communi y s uc u e di ec ly om ou da ase ins ead o elying on p e ious pa i ions
ha ha e been de i ed om adul connec omes (I o e al., 2017;Powe e al., 2011), because b ain connec i i y o adolescen s and
adul s is known o di e (Fai e al., 2009).
To p oduce he pa i ion, we a e aged he baseline sFC ma ices ac oss pa icipan s and emo ed nega i e en ies. The esul ing
ma ix was submi ed o he Lou ain communi y de ec ion algo i hm o weigh ed g aphs (Blondel e al., 2008) and his pa i ion was
e ined using a modula i y ine- uning algo i hm (Sun e al., 2009). Since he algo i hm is no de e minis ic, i was applied 100 imes
and he esul s ga he ed in a nodes x nodes consensus ma ix ha indica es he equency by which he co esponding node pai was
assigned o he same module. The consensus ma ix was pa i ioned epea edly un il con e gence. The algo i hm depends on a
pa ame e g ha con ols he esolu ion (which de e mines he ensuing numbe o modules). We adjus ed his pa ame e o maximize
he no malized mu ual in o ma ion be ween solu ions a di e en esolu ions. The op imal alue o gensu es he mos s able pa i-
ioning and in ou da ase (g= 2.7) led o a solu ion wi h 14 modules, a numbe ha yielded in e p e able modules and is on pa wi h
he ca dinali y used in p e ious s udies. These analyses a e simila o hose epo ed in (Gee ligs e al., 2015) and we e pe o med wi h
he B ain Connec i i y Toolbox (Rubino and Spo ns, 2010;h p://www.b ain-connec i i y- oolbox.ne ) o MATLAB. Ha ing pa cel-
la ed he connec ome in he 14 modules, we ained he p edic ion model o each one o hem using only connec ions implica ing
nodes in ha module (i. e. ei he connec ions among nodes in he module o connec ions be ween nodes in he module and he es o
he b ain). We employed he same module decomposi ion in he analysis conce ning he ollow-up da ase .
QUANTIFICATION AND STATISTICAL ANALYSIS
De i a ion, alida ion and psychome ic co ela es o Decision Acui y
We ailo ed analysis o es he hypo hesis ha a ew (a ound h ee) dimensions o co a ia ion would meaning ully load ac oss de-
cision-making measu es, expec ing ewa d sensi i i y, isk p e e ences, goal-di ec edness and p osociali y o be ep esen ed in
hese dimensions. We allowed, howe e , he da a o de e mine he numbe o ac o s in he model. We used an explo a o y-con i -
ma o y app oach o es ablish he s uc u e o he ac o model using he baseline da a. Then, we made use o he longi udinal na u e
o ou sample o es he empo al s abili y and p edic i e alidi y o he key de i ed measu e.
Task measu es a baseline only we e i s ans o med o nea -no mal ma ginal dis ibu ions using loga i hmic o powe -law ans-
o ms, impu ed o he small pe cen age o missing alues using he R package ‘missMDA’, hen andomly di ided in o a ‘disco e y’
and ‘ es ing’ samples. N = 416 pa icipan s we e used o explo a o y common ac o analysis (ECFA) and 414 we e used o ou -o -
sample es ing. We ound loadings on he i s ECFA ac o , likely o be mos impo an , o a y smoo hly ac oss all pa ame e s, and
he g ea majo i y o loadings o be lowe han he con en ional h eshold o 0.4 used o cons uc s uc u al equa ion models o
con i ma o y FA (Mu he
´n and Mu he
´n, 2008). I ems had high uniqueness, as expec ed. These esul s we e much like he inal o-
al-sample FA illus a ed in Figu e 1. The e o e, a he han claim ha ce ain decision pa ame e s we e impo an and o he s
we e no in p o iding a measu e o he unde lying la en a iable, we allowed o all decision-making i ems o con ibu e, ecognizing
ha indi idual i em weigh s would be poo ly es ima ed, bu expec ing ha he esul ing o e all sco es would be well es ima ed. We
es ed his by compa ing (i) disco e y e sus es samples and (ii) pu pose ul hal -spli s o he popula ion wi h espec o sex and age
(see supplemen al in o ma ion, sec ion B). The explo a o y analysis u he mo e sugges ed ha ou objec i e need no be o de e -
mine a g ound- u h numbe o ac o s, as highe o de ac o s we e domina ed by single asks and hence we e o no in e es he e.
Ou c i e ion o including highe -o de ac o s hen was whe he highe dimensional models we e likely o esul in be e sco e es-
ima es o he low-o de ac o s, which we e o in e es .
The maximum numbe o ac o s o he explo a o y-con i ma o y analysis was 8, es ima ed by pa allel analysis (Figu e S1). The R
lib a y ‘nFac o s’, and speci ically R unc ions ‘eigen’,’pa allel’ and ‘nSc ee’ we e used o es ima e he sc ee-plo based numbe o
explo a o y ac o s o e ain illus a ed in Figu e S1. Based on hese, models om 8 down o 1 ac o s we e de i ed wi h unc ion ‘ a’,
using o dina y leas -squa es o ind minimum- esidual (min es) solu ions. The models es ima ed om he ECFA we e hen es ed on
he con i ma ion da ase using s uc u al equa ion modeling in R (Fox, 2006). C i e ia o unde -de e mina ion, Bayesian In o ma ion
C i e ion (BIC), and compa a i e i index (CFI) we e used o compa e including an inc easing numbe o ac o s. In he con i ma o y
ac o analysis, a h eshold o 0.25 was adop ed as e y ew loadings on he i s ac o exceeded he con en ional h eshold o 0.4
(See Figu e 1 in he main ex ). Conside ing he es se o 414 pa icipan s only, we ound ha model i as indexed by he BIC and CFI
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040.e1–e7, June 16, 2021 e5
imp o ed om 1 o 4 ac o s. Howe e , 5 ac o and mo e complex model i s did no con e ge on he es se . Acco ding o hese
c i e ia we conside ed a model o ou ac o s o be mos pa simonious and obus , bu we also conside ed he s abili y o he la en
cons uc s de i ed o make a inal choice o ac o -analy ic model. Wi hin he ange o h ee o i e ac o s, dsco es we e no sensi i e
o he exac numbe o ac o s, sco es being co ela ed wi h > 0.9, p 0, wi h he sco e ob ained om he 4- ac o solu ion. We hus
op ed o a 4- ac o model o all subsequen analyses.
We hen es ed whe he decision acui y as a cons uc was s able wi h espec o (i) he andom disco e y/con i ma ion spli (ii)
median-spli age and (iii) sex using he baseline da a. We examined how closely sco es o a ce ain subg oup (below median age
o (ii), ‘Male’ o (iii)) based on ECFA o he g oup i sel ag eed wi h sco es o he same indi iduals based on FA weigh s de i ed
om he opposi e (i.e., abo e median age o ‘Female’) g oup. We examined he cons uc s abili y o decision acui y by co ela ing
componen dsco es on hal he sample wi h he same sco es de i ed om he i s ECFA componen on he o he hal o he sample.
He e we we e no p ima ily in e es ed in he ac o s uc u e o decision-making, bu in he s abili y o he cons uc o decision acui y.
We hus di ided he sample in o wo subg oups ei he by age (a 19) o by sex. We a gued ha i he cons uc i sel was s able ac oss
age (and sex), hen he decision acui y ac o sco e o each pa icipan could be calcula ed ei he using he ac o loadings de i ed
om he pa icipan ’s own g oup o indeed he opposi e one. Indi iduals wi h subs an ially di e ing sco es would indica e ha a
di e en la en cons uc o ganized decision-making ac oss he subg oups. I , o example, dwas an in a ian la en cons uc
wi h espec o age, hen he pa e n o loadings de i ed om olde pa icipan s would gi e he same sco es when applied o younge
pa icipan s as and ECFA on he young pa icipan da a hemsel es (Figu e S3). dwas highly s able ac oss he disco e y-con i ma-
ion andom spli (0.99 con idence in e al o (explo a o y based on con i ma o y loadings, own explo a o y) = 0.976,0.985), as well
as age CI (young|old, own young) = 0.969,0.9811). I s s abili y ac oss gende was sa is ac o y bu signi ican ly lowe , e idencing a
small deg ee o sexual dimo phism CI (male| emale, own male) = 0.820,0.887. Fi indica o s we e simila o he whole sample and o
each spli (e.g., RMSEA 90% CIs o emales, males, younge , olde and all we e 0.051-0.061, 0.052-0.062, 0.054-0.064, 0.046-0.056
and 0.052-0.058 espec i ely). None o he analyses was ma e ially a ec ed by excluding om he sample o 830 pa icipan s he 50
who had a diagnosis o DSM5 dep ession.
Finally, we es ed o ex e nal alidi y o decision acui y in co ela ing wi h (i ) men al heal h sco es o symp oma ology and dis-
posi ions, using bi ac o sco es and ( ) pa e ns o unc ional b ain connec i i y, as desc ibed in Resul s.
The ollow-up ba e y did no con ain one o he baseline asks, and had mino di e ences (bu he same de i ed pa ame e s) o
wo u he asks. In o de o pe o m longi udinal analyses, we adop ed a conse a i e app oach, es ima ing a measu e o decision
acui y based on he inal s age o he baseline analysis, bu e aining only he weigh s o he six asks ha we e assessed longi u-
dinally. We checked ha his mo e app oxima e measu e adequa ely cap u ed indi idual a iabili y o he baseline sample, which was
he case ( = 0.98, p unde ec able) and he e o e used in he longi udinal analysis baseline sco es de i ed om hese six asks. We
hen de i ed he ollow-up decision acui y es ima es as ollows. We i s applied he same app oxima e-gaussianiza ion ans o ms o
each ollow-up measu e. Nex , we z-sco ed each ollow-up measu e using he mean and s anda d de ia ion o he espec i e ( ans-
o med) baseline measu e. Finally, we applied he weigh s o hese 6 asks de i ed om he baseline ac o analysis. Thus, we ook
he ollow-up measu es o decision acui y o ha e exac ly he same s uc u e as baseline, so ha i could be used o compa e ab-
solu e changes in his measu e. Finally, in analyses co ela ing ollow-up symp oms wi h decision acui y, and as decision acui y and
IQ we e measu ed ypically six mon hs a e symp oms and hypo hesized o be ai -like, we in e pola ed ollow-up decision acui y
and IQ measu es o he ime o symp om measu emen .
Fo he longi udinal analysis, we used a linea mixed e ec s app oach. As he s uc u e o decision Acui y was ixed by he p o-
cedu e abo e, we did no spli he ollow-up sample in o es and disco e y se s. We used Bayesian In o ma ion C i e ion o selec
he s a is ical models by which we es ed o in e - ela ions be ween decision acui y and key psychome ic a iables. Fo all he
ollowing analyses, N = 571 o he ollow up sample. De elopmen al ime in his accele a ed longi udinal design is ep esen ed
bo h by age-a - ec ui men , and by he ime in e al be ween es wa es. Bo h ec ui men p ocedu es and de elopmen i sel
may mean ha hese wo measu es o de elopmen al age may in p ac ice a ec ou dependen a iables di e en ly. We he e o e
i s checked i LME modeling o e baseline and ollow-up wi h age as a andom e ec , in addi ion o a andom in e cep o each
pa icipan , imp o ed model i . In ac , i wo sened model i (BIC = 5974.5; logLik = 2965.512; e sus BIC = 5960.0, logLik =
2965.529), so we did no include age as andom e ec in u he analyses. In u he analyses in ol ing IQ, we used he aw ma ix
and ocabula y WASI IQ subsco es and modeled age explici ly, a he han use s anda dized IQ subsco es. This is because we
no iced ha he s anda dized WASI o al IQ in ou sample was associa ed wi h age ( Pea son = 0.135, p = 0.00011,
2
= 0.017) a
baseline. This indica es ha ou sample had a di e en age dependence o IQ sco es han he e e ence one (Axel od, 2002). The e-
o e, we eg essed d o aw IQ subsco es while co a ying o age, in e ec accoun ing o a ia ion in IQ abili y independen o
whe he his was due o age o sel -selec ion.
We also o med a pe o mance measu e ac oss asks, in o de o check he in e p e a ion ha d e lec s be e decision-making.
Fi s , we excluded he discoun ing and Roule e asks, as hese speci ically p obed he balance o amoun s won e sus o he dimen-
sions o he e u n, namely i s delay and unce ain y espec i ely. Second, we excluded he App oach-A oidance con lic ask, as one
o he measu es by which i en e ed he es ima ion o dwas judged o be oo close o a pe o mance measu e al eady ( he hi d com-
mon- ac o sco es o he wi hin- ask ac o analysis; see abo e). Pa en he ically, his measu e loaded modes ly in he expec ed di-
ec ion on o d, i.e., posi i ely, wi h a weigh o 0.24 and high uniqueness (Figu e 1). We hen -sco ed winnings wi hin each o he
Go-NoGo, In o ma ion Ga he ing, In es o -T us ee and Two-s ep asks, and a e aged hese sco es ac oss asks. The Pea son
ll
OPEN ACCESS A icle
e6 Neu on 109, 2025–2040.e1–e7, June 16, 2021
aw and pa ial co ela ion able be ween his ask-pe o mance measu e, dand WASI o al IQ, shown in Table S2, suppo ed he
in e p e a ion o das conduci e o p o i able decision-making abo e and beyond IQ.
P edic i e pe o mance o Connec i i y Analysis
We assessed p edic i e pe o mance as he Pea son co ela ion coe icien be ween measu ed dand (c oss- alida ed) p edic ed
d(d
p
), a e aged ac oss epe i ions o he c oss- alida ion spli s. A e Fishe ans o ma ion, he null dis ibu ion o should ollow a
ze o-cen e ed Gaussian dis ibu ion. In o de o app aise signi icance, we es ima ed he a iance o his dis ibu ion by gene a ing
100 andom pe mu a ions o he a ge a iable (Winkle e al., 2016) and epea ing he model- i ing p ocedu es men ioned abo e,
sepa a ely o each old. We hen de i ed p alues o he obse ed om he es ima ed null dis ibu ion. We assessed p edic i e
pe o mance o a model based on he ull se o connec ions, as well as o models ained on he subse s o connec ions co e-
sponding o he modules desc ibed in he p e ious subsec ion.
To demons a e ha he ela ionships be ween connec i i y and decision acui y we e s able o e ime and eplica e, we used he
model es ima ed a baseline o p edic dbased on he ollow-up sFC da a o modules ha we e signi ican a baseline. Gi en ha
he da a a baseline and ollow-up a e no independen , we kep he same c oss- alida ion old s uc u e in bo h da ase s, so ha he
p edic ion o a subjec a ollow-up did no in ol e hei own sFC baseline da a, as his would ha e in la ed he es ima es o p edic i e
pe o mance a ollow-up.
Connec i i y pa e ns p edic i e o d e sus IQ
Fo imaging analyses, we de i ed a composi e sco e o IQ by a e aging s anda dized ocabula y and ma ix IQ subsco es, a he
han using he s anda dized WASI sco e, because o wo easons. Fi s , we wan ed analyses in ol ing bo h age and IQ o ha e a
s aigh o wa d in e p e a ion whe e IQ ep esen s a measu e o aw abili y, as opposed o age-s anda dized abili y, and explici ly
es o age-dependence sepa a ely. Second, we ound e idence (Resul s) ha ou sample was di e en om he o iginal on which
s anda dized sco es we e de i ed, and hence he s anda diza ion p ocedu e migh be in alid. Nex , we ained models bo h on he
comple e se o connec ions and he subse s co esponding o he indi idual modules o p edic he IQ composi e sco es, as we had
done p e iously o p edic d, yielding IQ
p
, and assessed p edic i e pe o mance o each o he modules sepa a ely. To compa e he
connec i i y pa e ns ha we e p edic i e o dwi h hose p edic i e o IQ, o each o he modules we assessed he pa ial co ela ion
be ween dand d
p
when con olling o IQ, and he pa ial co ela ion be ween IQ and IQ
p
when con olling o d. In all hese analyses
we co ec ed o age, sex and imaging- ela ed con ounds as abo e.
ll
OPEN ACCESS
A icle
Neu on 109, 2025–2040.e1–e7, June 16, 2021 e7