This is a sel -a chi ed e sion o an o iginal a icle. This e sion
may di e om he o iginal in pagina ion and ypog aphic de ails.
Au ho (s):
Ti le:
Yea :
Ve sion:
Copy igh :
Righ s:
Righ s u l:
Please ci e he o iginal e sion:
CC BY 4.0
h ps://c ea i ecommons.o g/licenses/by/4.0/
Genome-wide analyses o indi idual di e ences in quan i a i ely assessed eading- and
language- ela ed skills in up o 34,000 people
© 2022 he Au ho (s). Published by PNAS
Published e sion
Eising, Else; Mi za-Sch eibe , Nazanin; de Zeeuw, E eline L.; Wang, Ca ol A.;
T uong, Dongnhu T.; Alleg ini, And ea G.; Shapland, Chin Yang; Zhu, Gu; Wigg,
Ka en G.; Ge i se, Ma go L.; Molz, Ba ba a; Alagöz, Gökbe k; Gialluisi,
Alessand o; Abbondanza, Filippo; Rim eld, Kaili; an Donkelaa , Ma jolein; Liao,
Zhijie; Jansen, Philip R.; Andlaue , Till F. M.; Ba es, Timo hy C.; Be na d, Manon;
Blokland, Ki s en; Bon e, Milene; Bø glum, Ande s D.; Bou ge on, Thomas;
B andeis, Daniel; Ce oni, Fabiola; Csépe, Valé ia; Dale, Philip S.; de Jong, Pe e F.;
DeF ies, John C.; Démone , Jean-F ançois; Demon is, Di e; Feng, Yu; Go don,
Sco D.; Guge , Sha on L.; Hayiou-Thomas, Ma ianna E.; He nández-Cab e a,
Juan A.; Ho enga, Jouke-Jan; Hulme, Cha les; Ke e, Juha; Ke , Elizabe h N.;
Kooma , Tanne ; Lande l, Ka in; Leona d, Gab iel T.; Lo e , Mau een W.;
Lyy inen, Heikki; Ma in, Nicholas G.; Ma inelli, Angela; Mau e , U s;
Michaelson, Jacob J.; Moll, K is ina; Monaco, An hony P.; Mo gan, Angela T.;
Nö hen, Ma kus M.; Pauso a, Zdenka; Pennell, C aig E.; Penning on, B uce F.;
P ice, Kai lyn M.; Rajagopal, Vee a M.; Ramus, F anck; Riche , Louis; Simpson,
Nuala H.; Smi h, Shelley D.; Snowling, Ma ga e J.; S ein, John; S ug, Lisa J.;
Talco , Joel B.; Tiemeie , Henning; an de Sch oe , Ma c P.; Ve hoe , Ellen;
Wa kins, Ka e E.; Wilkinson, Ma ga e ; W igh , Ma ga e J.; Ba , Ca hy L.;
Boomsma, Do e I.; Ca ei as, Manuel; F anken, Ma ie-Ch is ine J.; G uen,
Je ey R.; Luciano, Michelle; Mülle -Myhsok, Be am; Newbu y, Dianne F.;
Olson, Richa d K.; Pa acchini, Sil ia; Paus, Tomáš; Plomin, Robe ; Reilly, Sheena;
Schul e-Kö ne, Ge d; Tomblin, J. B uce; an Be gen, Elsje; Whi ehouse, And ew J.
O.; Willcu , E ik G.; S Pou cain, Bea e; F ancks, Clyde; Fishe , Simon E.
Eising, E., Mi za-Sch eibe , N., de Zeeuw, E. L., Wang, C. A., T uong, D. T., Alleg ini, A. G.,
Shapland, C. Y., Zhu, G., Wigg, K. G., Ge i se, M. L., Molz, B., Alagöz, G., Gialluisi, A.,
Abbondanza, F., Rim eld, K., an Donkelaa , M., Liao, Z., Jansen, P. R., Andlaue , T. F. M., . . .
Fishe , S. E. (2022). Genome-wide analyses o indi idual di e ences in quan i a i ely assessed
eading- and language- ela ed skills in up o 34,000 people. P oceedings o he Na ional
Academy o Sciences o he Uni ed S a es o Ame ica, 119(35), A icle e2202764119.
h ps://doi.o g/10.1073/pnas.2202764119
2022
Genome-wide analyses o indi idual di e ences in
quan i a i ely assessed eading- and language- ela ed
skills in up o 34,000 people
Else Eising
a
, Nazanin Mi za-Sch eibe
b
, E eline L. de Zeeuw
c
, Ca ol A. Wang
d,e
, Dongnhu T. T uong
, And ea G. Alleg ini
g
, Chin Yang Shapland
h,i
,
Gu Zhu
j
, Ka en G. Wigg
k
, Ma go L. Ge i se
a
, Ba ba a Molz
a
,G
€
okbe k Alag€
oz
a
, Alessand o Gialluisi
l,m,n
, Filippo Abbondanza
o
, Kaili Rim eld
g,p
,
Ma jolein an Donkelaa
a
, Zhijie Liao (廖志洁)
q
, Philip R. Jansen
,s,
, Till F. M. Andlaue
l,u
, Timo hy C. Ba es
, Manon Be na d
w
, Ki s en Blokland
x
,
Milene Bon e
y
, Ande s D. Bø glum
z,aa,bb
, Thomas Bou ge on
cc
, Daniel B andeis
dd,ee, ,gg
, Fabiola Ce oni
hh,ii
,Val

e ia Cs
epe
jj,kk
, Philip S. Dale
ll
,
Pe e F. de Jong
mm
, John C. DeF ies
nn,oo
, Jean-F anc¸ois D
emone
pp
, Di e Demon is
z,aa
, Yu Feng
k
, Sco D. Go don
j
, Sha on L. Guge
qq
,
Ma ianna E. Hayiou-Thomas
, Juan A. He n
andez-Cab e a
ss
, Jouke-Jan Ho enga
c
, Cha les Hulme
, Juha Ke e
uu,
, Elizabe h N. Ke
qq,ww,xx
,
Tanne Kooma
yy
, Ka in Lande l
zz,aaa
, Gab iel T. Leona d
bbb
, Mau een W. Lo e
x,xx
, Heikki Lyy inen
ccc
, Nicholas G. Ma in
j
, Angela Ma inelli
o
,
U s Mau e
ddd
, Jacob J. Michaelson
yy
, K is ina Moll
eee
, An hony P. Monaco
, Angela T. Mo gan
ggg,hhh,iii
, Ma kus M. N€
o hen
jjj
, Zdenka Pauso a
w,kkk
,
C aig E. Pennell
d,e,lll
, B uce F. Penning on
mmm
, Kai lyn M. P ice
k,x,nnn
, Vee a M. Rajagopal
z,aa
, F anck Ramus
ooo
, Louis Riche
ppp
,
Nuala H. Simpson
qqq
, Shelley D. Smi h
, Ma ga e J. Snowling
qqq,sss
, John S ein
, Lisa J. S ug
uuu,
, Joel B. Talco
www
, Henning Tiemeie
,xxx
,
Ma c P. an de Sch oe
yyy,zzz
, Ellen Ve hoe
a
, Ka e E. Wa kins
qqq
, Ma ga e Wilkinson
x
, Ma ga e J. W igh
aaaa
, Ca hy L. Ba
k,x,nnn
,
Do e I. Boomsma
c,bbbb,cccc
, Manuel Ca ei as
dddd,eeee,
, Ma ie-Ch is ine J. F anken
yyy
, Je ey R. G uen
, Michelle Luciano
,
Be am M€
ulle -Myhsok
l,gggg
, Dianne F. Newbu y
ii
, Richa d K. Olson
nn
, Sil ia Pa acchini
o
,Tom

a
s Paus
hhhh
, Robe Plomin
g
, Sheena Reilly
ggg,iiii
,
Ge d Schul e-K€
o ne
eee
, J. B uce Tomblin
jjjj
,Elsje anBe gen
c,bbbb,kkkk
, And ew J. O. Whi ehouse
llll
, E ik G. Willcu
oo
,Bea eS Pou cain
a,h,mmmm,1
,
Clyde F ancks
a,mmmm,nnnn,1
, and Simon E. Fishe
a,mmmm,2
Edi ed by U a F i h, Uni e si y College London, London, Uni ed Kingdom; ecei ed Feb ua y 18, 2022; accep ed May 31, 2022
The use o spoken and w i en language is a undamen al human capaci y. Indi idual di -
e ences in eading- and language- ela ed skills a e influenced by gene ic a ia ion, wi h
win-based he i abili y es ima es o 30 o 80% depending on he ai . The gene ic a chi-
ec u e is complex, he e ogeneous, and mul i ac o ial, bu in es iga ions o con ibu ions
o single-nucleo ide polymo phisms (SNPs) we e hus a unde powe ed. We p esen a
mul icoho genome-wide associa ion s udy (GWAS) o fi e ai s assessed indi idually
using psychome ic measu es (wo d eading, nonwo d eading, spelling, phoneme awa e-
ness, and nonwo d epe i ion) in samples o 13,633 o 33,959 pa icipan s aged 5 o 26 y.
We iden ified genome-wide significan associa ion wi h wo d eading ( s11208009,
P=1.098 ×10
28
) a a locus ha has no been associa ed wi h in elligence o educa ional
a ainmen . All fi e eading-/language- ela ed ai s showed obus SNP he i abili y,
accoun ing o 13 o 26% o ai a iabili y. Genomic s uc u al equa ion modeling
e ealed a sha ed gene ic ac o explaining mos o he a ia ion in wo d/nonwo d eading,
spelling, and phoneme awa eness, which only pa ially o e lapped wi h gene ic a ia ion
con ibu ing o nonwo d epe i ion, in elligence, and educa ional a ainmen . A mul i a i-
a e GWAS o wo d/nonwo d eading, spelling, and phoneme awa eness maximized powe
o ollow-up in es iga ion. Gene ic co ela ion analysis wi h neu oimaging ai s iden i-
fied an associa ion wi h he su ace a ea o he banks o he le supe io empo al sulcus, a
b ain egion linked o he p ocessing o spoken and w i en language. He i abili y was
en iched o genomic elemen s egula ing gene exp ession in he e al b ain and in ch o-
mosomal egions ha a e deple ed o Neande hal a ian s. Toge he , hese esul s p o ide
a enues o deciphe ing he biological unde pinnings o uniquely human ai s.
eading jlanguage jgenome-wide associa ion s udy jme a-analysis
The p ocessing and p oduc ion o complex spoken and w i en language a e capaci ies
ha appea o be dis inc o ou species (1). Such skills ha e become undamen al o
day- o-day li e in mode n socie y. Decades o amily and win s udies ha e e ealed
subs an ial gene ic componen s con ibu ing o indi idual a ia ion in eading- and
language- ela ed ai s as well as o suscep ibili y o associa ed diso de s (2). A ecen
me a-analysis in eg a ed a ailable da a on hese skills om 49 win s udies, wi h a o al
sample size o 38,000 child en and adolescen s aged 4 o 18 y. The me a-analysis
yielded he i abili y es ima es o 66% o wo d eading (me a-analysis o 48 s udies),
80% o spelling (15 s udies), and 52% o phoneme awa eness ( he abili y o iden i y
and manipula e indi idual sounds o spoken wo ds; 13 s udies) and sugges ed g ea e
gene ic influences on eading- ela ed abili ies han language- ela ed measu es ( win he -
i abili y o 34%; me a-analysis o 10 s udies wi h measu es on ecep i e and exp essi e
ocabula y, o al language, and naming abili ies) (3).
Significance
Ou unique capaci ies o spoken
and w i en language a e
undamen al ea u es o wha
makes us human, ye he
biological bases emain la gely
mys e ious. We p esen a la ge-
scale well-powe ed genome-wide
associa ion s udy me a-analysis o
indi idual di e ences in eading-
and language- ela ed skills (wo d
eading, nonwo d eading,
spelling, phoneme awa eness,
and nonwo d epe i ion) in ens o
housands o pa icipan s. The
findings p omp a majo
ee alua ion o p io li e a u e
claiming candida e gene
associa ions in much smalle
samples. Mo eo e , we use he
no el gene ic da a as windows
in o mul iple aspec s o he
biology o hese impo an
abili ies, e ealing molecula links
o indi idual di e ences in
neu oana omy o language-
ela ed b ain a eas and en iched
he i abili y in a chaic dese s o
he human genome as well as in
e al b ain enhance egions.
This a icle is a PNAS Di ec Submission.
Copy igh © 2022 he Au ho (s). Published by PNAS.
This open access a icle is dis ibu ed unde C ea i e
Commons A ibu ion License 4.0 (CC BY).
Published Augus 23, 2022.
PNAS 2022 Vol. 119 No. 35 e2202764119 h ps://doi.o g/10.1073/pnas.2202764119 1o 12
RESEARCH ARTICLE
|
PSYCHOLOGICAL AND COGNITIVE SCIENCES
GENETICS OPEN ACCESS
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.
Linkage mapping and a ge ed candida e gene s udies ha e
epo ed associa ions o SNPs and/o gene ic loci wi h eading-
and language- ela ed ai s as well as wi h diso de s such as dyslexia
and de elopmen al language diso de (DLD), which encompasses
he olde defini ion o specific language impai men (SLI) (4).
Howe e , eplica ion e o s ha e been me wi h limi ed success
(4). Mo eo e , o he language sciences, in con as o o he a eas
o human gene ics, he e ha e so a been ew genome-wide associ-
a ion s udies (GWASs), in which SNPs a millions o poin s ac oss
he genome a e sys ema ically sc eened o associa ion wi h he ai
o in e es in la ge da ase s comp ising housands o indi iduals
(2). GWAS e o s a e beginning o iden i y SNPs ha show
genome-wide significan associa ions wi h eading- and language-
ela ed ai s: s7642482 nea ROBO2 associa ed wi h exp essi e
ocabula y in in ancy (5); s17663182 wi hin MIR924HG wi h
apid au oma ized naming o le e s (6); and s1555839 nea
RPL7P34 wi h apid au oma ized naming and apid al e na ing
s imulus, defici s o which a e o en implica ed in dyslexia (7).
None heless, insigh s in o he genomic unde pinnings o hese
ypes o skills om GWAS app oaches ha e hus a been limi ed,
which may eflec low powe due o he ela i ely small sample
sizes o he coho s, such ha he majo i y o gene ic a iance
emains unexplained. Sample sizes ha e emained limi ed because
o he labo -in ensi e assessmen me hods equi ed o pheno ypic
cha ac e iza ion o eading- and language- ela ed ai s, which a e
di ficul o e en impossible o eplace wi h simple ques ionnai es.
Ye , well-powe ed GWAS e o s ha cha ac e ize he molecula
gene ic a ia ion in ol ed in eading- and language- ela ed ai s
ha e he po en ial o p o ide no el pe spec i es on he biological
bases and o igins o human cogni i e specializa ions (8).
He e, we p esen la ge-scale GWAS me a-analyses o a se o
eading- and language- ela ed ai s, measu ed wi h psychome ic
ools. We cap u ed a ia ion ac oss he pheno ypic spec um,
ex ending beyond diso de . Ou s udy ocused on ai s assessed
using con inuous measu es in mul iple coho s om he in e na-
ional GenLang ne wo k (h ps://www.genlang.o g/) oge he
wi h se e al public da ase s ha ha e da a a ailable o he ele an
pheno ypes ma ched o genome-wide geno ype in o ma ion. Fi e
quan i a i e ai s we e iden ified o which pheno ype da a could
be aligned ac oss di e en coho s o yield su ficien ly la ge sample
sizes o GWAS: wo d eading, nonwo d eading, spelling, pho-
neme awa eness, and nonwo d epe i ion. Uni a ia e GWAS
me a-analyses we e pe o med o each o he pheno ypes o iden-
i y gene ic a ia ion influencing hese ai s and o model gene ic
o e laps be ween hem. Fo compa a i e pu poses, a GWAS me a-
analysis o pe o mance in elligence quo ien (IQ) was also pe -
o med in he same da ase . Toge he wi h publicly a ailable
GWAS summa y s a is ics om p io s udies o cogni i e pe o -
mance and educa ional a ainmen , hese da a we e used o s udy
gene ic ela ionships be ween eading- and language- ela ed ai s,
IQ, and educa ional a ainmen . A mul i a ia e app oach allowed
us o op imize he powe o GWAS me a-analysis o unc ional
ollow-ups, gi ing insigh s in o he issues, cell ypes, b ain egions,
and e olu iona y signa u es in ol ed.
Resul s
Me a-Analyses o Quan i a i e Reading- and Language-Rela ed
T ai s in 22 Coho s. Ou s udy ocused on fi e quan i a i e ead-
ing- and language- ela ed ai s: wo d eading accu acy, nonwo d
eading accu acy, spelling accu acy, phoneme awa eness, and
nonwo d epe i ion accu acy (Table 1). These ai s a e hough
o ap in o a numbe o unde lying p ocesses in ol ed in w i en
and spoken language. Fo example, nonwo d eading elies
hea ily on basic decoding skills [ ansla ing g aphemes one by
one in o phonemes (9)], while spelling u ilizes lexical and o ho-
g aphic knowledge [unde s anding o pe missible le e pa e ns
and how hey a e a anged in specific wo ds (10)]. Phoneme
awa eness measu es he abili y o dis inguish and manipula e he
sepa a e phonemes in spoken wo ds (11). Nonwo d epe i ion
asks ap in o speech pe cep ion, phonological sho - e m mem-
o y, and a icula ion (12). A o al o 22 coho s agg ega ed by he
GenLang Conso ium combined wi h se e al publicly a ailable
da ase s p o ided da a o one o se e al o hese ai s (Da ase s
S1–S3 and SI Appendix,Figs.S1andS2). The coho s connec ed
in he GenLang ne wo k ei he we e o iginally asce ained h ough
a p oband wi h a language/ eading diso de (DLD/SLI o dyslexia)
o we e sampled om he gene al popula ion; all coho s include
quan i a i e pheno ypic da a ga he ed ia alida ed psychome ic
es s as well as genome-wide geno ype da a om he es ed indi id-
uals. Some o he samples a e bi h coho s, and some in ol e am-
ily o win designs. The pheno ype da a we e collec ed ac oss an
a ay o di e en ages, es ins umen s, and languages (p ima ily
English bu also, Du ch, Spanish, Ge man, F ench, Finnish, and
Hunga ian). We educed he e ogenei y o assessmen age by
excluding indi iduals o e 18 y o age (excep o h ee coho s)
(SI Appendix,Ex ended Me hods) and whe e pheno ype da a we e
a ailable om he same pa icipan a mul iple ages, by choosing
Table 1. Pheno ypes and sample sizes o he GWAS me a-analyses
T ai Pheno ype desc ip ion
Me a-analysis
o al sample
Me a-analysis
Eu opean ances y only
No. o
coho s
No. o
indi iduals
No. o
coho s
No. o
indi iduals
Wo d eading Numbe o co ec wo ds ead aloud om a lis
in a ime- es ic ed o un es ic ed ashion
19 33,959 18 27,180
Nonwo d* eading Numbe o nonwo ds ead aloud co ec ly om a
lis in a ime- es ic ed o un es ic ed ashion
13 17,984 12 16,746
Spelling Numbe o wo ds co ec ly spelled o ally o in
w i ing a e being dic a ed as single wo ds o
in a sen ence
15 18,514 14 17,278
Phoneme awa eness Numbe o wo ds co ec ly al e ed in phoneme
dele ion/elision and spoone ism asks
12 13,633 11 12,411
Nonwo d* epe i ion Numbe o nonwo ds o phonemes epea ed
aloud co ec ly
10 14,046 10 12,828
*A nonwo d is a g oup o phonemes ha looks o sounds like a wo d, obeys he phono ac ic ules o he language, bu has no meaning.
2o 12 h ps://doi.o g/10.1073/pnas.2202764119 pnas.o g
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.
heage ha ma chedbes wi h heassessmen ageso hela ges
coho (s). We limi ed he he e ogenei y in oduced by di e en
es ins umen s by only including hose ha measu ed he pheno-
ypes desc ibed in ou analysis plan (SI Appendix,Supplemen al
No es) and in cases whe e da a om mo e han one es ins u-
men we e a ailable, by selec ing he es ins umen ha was used
by he la ges numbe o coho s.
We e alua ed whe he , despi e e o s o minimize his, he -
e ogenei y ela ed o age and/o use o di e en es ins umen s
emained in ou da a. To do so, we gene a ed GWAS me a-
analysis esul s o wo d and nonwo d eading s a ified by age o
es ins umen (Da ase S4). We es ed hose da a o gene ic co -
ela ions using linkage disequilib ium sco e eg ession (LDSC)
(13), de e mining o wha ex en he same common gene ic a ia-
ion is accoun ing o pheno ypic a iabili y in he di e en s a i-
fied GWAS da ase s (de ailed in SI Appendix, Supplemen al No es).
The e was limi ed he e ogenei y o GWAS me a-analysis esul s, as
also e iden in he Coch an Q s a is ics (SI Appendix,Fig.S3)and
LDSC a ios (Da ase S4) o all ai s excep nonwo d epe i ion.
We also assessed e ec s o sex, obse ing high gene ic co ela ions
o emale- and male-only subse s (SI Appendix Supplemen al No es,
Da ase S4). Gi en he lack o gene ic he e ogenei y, he emainde
o he s udy in ol ed analyses o he ull non-s a ified da ase , o
ensu e la ges a ailable sample size and maximal powe .
A Genome-Wide Signi ican Locus Associa ed wi h Wo d Reading.
We pe o med uni a ia e GWAS me a-analyses o he eading-/
language- ela ed ai s in he ull GenLang da ase as ollows. Fo
each pheno ype, associa ions be ween SNPs and he quan i a i e
ai we e calcula ed in e e y coho sepa a ely, hen combined
in o a me a-analysis o ha pheno ype. This yielded fi e sepa a e
uni a ia e GWAS da ase s, one o each ai . Fo e alua ing s a-
is ical significance o SNP associa ions, we de e mined an app o-
p ia e h eshold ha was adjus ed no only o genome-wide
sc eening (P=5×10
8
), bu also o mul iple es ing based
on he co ela ion s uc u e o ou fi e eading-/language-
ela ed ai s, as es ima ed using pheno ypic Spec al Decom-
posi ion (phenoSpD; Ma e ials and Me hods). The significance
h eshold o assessing he GWAS me a-analysis esul s was
hus se o P=5×10
8
/2.15 independen ai s =2.33 ×
10
8
. We iden ified a genome-wide significan locus associa ed
wi h wo d eading ( s11208009 C/T on ch omosome 1, P=
1.10 ×10
8
, be a =0.048, SE =0.008) (SI Appendix,Fig.
S5). No ably, s11208009 has no shown associa ion wi h gene al
cogni i e pe o mance o educa ional a ainmen , while o he
SNPs in linkage disequilib ium (LD) wi h s11208009 (
2
>0.6)
ha e been associa ed wi h iglyce ide and o al choles e ol le els
in blood in p e ious GWAS (Da ase S5). Th ee genes a e
loca ed in he icini y o s11208009 and SNPs in LD (
2
>
0.6): DOCK7,encodingaguaninenucleo ideexchange ac o
impo an o neu ogenesis (14); ANGPTL3,whichencodesa
g ow h ac o specific o he ascula endo helium ha is
exp essed specifically in he li e (15); and USP1,encodingadeu-
biqui ina ing enzyme specific o he Fanconi anemia pa hway
(16). The associa ed locus ha bo s an exp ession quan i a i e ai
locus egula ing DOCK7 and ATG4C [ano he nea by gene ha
encodes an au ophagy egula o (17)] in he ce ebellum and
DOCK7,ATG4C,andUSP1 in blood (Da ase S6). Genome-
wide significan loci we e no iden ified o he o he ai s.
Da ase S7 lis s all esul s wi h P<1×10
6
.
T ai s Rela ed o W i en and Spoken Language A e Highly
Co ela ed a he Gene ic Le el. The indi idual e ec size o
ou genome-wide significan hi is small, as is ypical o
gene ically complex ai s. We wen on o make use o he com-
ple e GWAS signal conside ed in agg ega e ac oss he genome
o gain insigh s in o he gene ic a chi ec u e o he eading-/
language- ela ed ai s as well as ela ionships wi h o he aspec s
o human biology. Fi s , o each pheno ype, we es ima ed
SNP-based he i abili y: he p opo ion o ai a iabili y
explained by he SNPs included in he GWAS. All fi e ai s
showed significan SNP-based he i abili y, wi h LDSC-based
es ima es anging om 0.13 o nonwo d epe i ion o 0.26 o
nonwo d eading (Da ase S4), indica ing ha he cap u ed
common gene ic a ia ion accoun s o a subs an i e p opo ion
o he pheno ypic a iance. These obse a ions allowed o
ollow-up analyses ha a e dependen on significan SNP he i a-
bili y, including es ima es o gene ic co ela ions: a measu e ha
quan ifies he o e all gene ic simila i y be ween wo complex
ai s. Pai wise gene ic co ela ion analyses showed significan
o e lap among he eading- and language- ela ed ai s (Fig. 1A
and Da ase S4). Gene ic co ela ion es ima es we e especially
high o wo d eading, nonwo d eading, spelling, and phoneme
awa eness, anging om 0.96 (SE =0.07) o 1.06 (SE =0.07).
P io li e a u e has shown pheno ypic co ela ions o eading-
and language- ela ed ai s wi h gene al cogni i e pe o mance
and educa ional a ainmen . Mos cogni i e assessmen s depend
on a combina ion o e bal and non e bal es s. To enable he
in es iga ion o gene ic o e laps be ween non e bal cogni i e
pe o mance and eading- and language- ela ed ai s while
closely ma ching he sample cha ac e is ics o ou s udy, we
ca ied ou a GWAS me a-analysis o pe o mance IQ in he
GenLang ne wo k (n=18,722) (SI Appendix, Figs. S1–S3).
Only non e bal sub es s o gene al in elligence es s we e used
in his analysis (Da ase S1). Summa y s a is ics we e also
ob ained om ano he h ee sou ces: 1) genome-wide s udies
o ull-scale IQ (based on bo h e bal and non e bal asks; n=
257,828) and educa ional a ainmen (n=766,345) by he
Social Science Gene ic Associa ion Conso ium (18); 2) a GWAS
by sub ac ion s udy ha in es iga ed he noncogni i e abili ies
in ol ed in educa ional a ainmen (n=510,795) (19); and 3) a
ecen GWAS analysis o school g ades in he Danish In eg a i e
Psychia ic Resea ch (iPSYCH) coho (n=30,982) ha used a
decomposi ion analysis o iden i y gene ic associa ions wi h dis-
inc domains o pe o mance (20).
The fi e GenLang eading-/language- ela ed ai s showed
mode a e o s ong posi i e gene ic co ela ions wi h ull-scale
IQ ( ange =0.52 o 0.77), educa ional a ainmen ( ange =
0.54 o 0.68), and school pe o mance ( ange =0.54 o 0.81)
(Fig. 1Aand Da ase S8). In e es ingly, gene ic co ela ions wi h
ull-scale IQ we e subs an ially highe o wo d eading (95% CI =
0.70 o 0.85) han nonwo d eading (95% CI =0.50 o 0.68),
likely eflec ing he impo ance o eading skills o e bal es s
o cogni ion. Gene ic co ela ions o eading-/language- ela ed
ai s wi h pe o mance IQ ( ange =0.20 o 0.35) we e much
lowe han hose o ull-scale IQ, and he 95% CIs did no
o e lap. Indeed, only wo d eading showed a significan gene ic
co ela ion wi h pe o mance IQ. These esul s indica e ha
eading-/language- ela ed ai s and IQ a e a leas pa ly based
on dis inc gene ic ac o s. Significan ai -specific gene ic co e-
la ions we e obse ed o componen s 2 o 4 o he Danish
school g ade decomposi ion analysis (20). Componen 2, eflec -
ing ela i ely be e school g ade pe o mance in language han
ma hema ics (as compa ed wi h pee s), was posi i ely co ela ed
wi h bo h GenLang eading ai s. Componen 3, eflec ing
ela i ely be e school g ade pe o mance in o al han in
w i en examina ions, showed significan nega i e co ela ions
wi h phoneme awa eness and spelling. Las ly, componen 4,
PNAS 2022 Vol. 119 No. 35 e2202764119 h ps://doi.o g/10.1073/pnas.2202764119 3o 12
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.
eflec ing ela i ely be e school g ade pe o mance in Danish
han in English, showed significan nega i e co ela ion wi h
nonwo d epe i ion. The noncogni i e abili ies in ol ed in edu-
ca ional a ainmen iden ified in he GWAS by sub ac ion s udy
(19) showed small bu significan posi i e gene ic co ela ions
wi h wo d eading, nonwo d eading, and nonwo d epe i ion.
Genomic s uc u al equa ion modeling (GenomicSEM) (21)
is a me hod ha can use GWAS esul s o fi and compa e
models desc ibing gene ic o e laps be ween mul iple pheno-
ypes o u he unde s and how ai s a e ela ed. We used his
me hod o model he sha ed gene ic a chi ec u e o he fi e
eading- and language- ela ed ai s oge he wi h pe o mance
IQ, ull-scale IQ, and educa ional a ainmen (Da ase S9).
Explo a o y ac o models wi h one o ou ac o s we e fi ed
o he da a, o which he h ee- ac o model explained he
majo i y o he a iance. The h ee- ac o model was ollowed
up using confi ma o y ac o analysis in GenomicSEM. In he
final model (Fig. 1B), he fi s ac o explains a ia ion in non-
wo d eading, spelling, phoneme awa eness, wo d eading, and
ull-scale IQ. Fo he fi s h ee ai s, he e is no e idence o
addi ional gene ic influences, sugges ing high gene ic simila i y.
The second ac o explains addi ional a ia ion in ull-scale IQ
and is also ela ed o pe o mance IQ and educa ional a ain-
men . The hi d ac o explains a ia ion in nonwo d epe i-
ion, wo d eading, and educa ional a ainmen . Fac o s 1 and
3 a e highly co ela ed, indica ing ha he gene ic a chi ec u e
unde lying wo d eading does no di e much om nonwo d
eading, spelling, and phoneme awa eness. Nonwo d epe i ion,
on he o he hand, is gene ically mo e dis inc , as indica ed by
e idence o specific gene ic influences no cap u ed by he
model. Specific gene ic influences we e also e iden o
ull-scale IQ and educa ional a ainmen . Thus, al hough ead-
ing- and language- ela ed ai s show gene ic o e laps wi h ull-
scale IQ and educa ional a ainmen , he model indica es ha
hese ai s also ha e unique unsha ed componen s, in line wi h
gene ic co ela ion es ima es ha a e lowe han one.
Limi ed E idence o Genes P e iously Repo ed in Reading-/
Language-Rela ed T ai s and Diso de s. The numbe o p e i-
ous GWAS on eading-/language- ela ed ai s and diso de s is
small (Da ase S10), and hese ha e iden ified e y ew associa-
ions exceeding genome-wide significance. In hose p io s udies, a
o al o 48 independen SNPs me a less s ingen h eshold o
P<1×10
6
in he espec i e GWAS. We an lookups o each
o hose SNPs in ou GenLang GWAS me a-analysis esul s
(Da ase S10). Whe e SNP associa ions passed a h eshold adjus ed
o mul iple es ing o 48 SNPs and 2.15 independen GenLang
ai s (P<4.84 ×10
4
), we hen e an he associa ion analyses
a e exclusion o he o iginal coho (s) in which he associa ion was
fi s iden ified o e alua e independen e ec s beyond hose o he
espec i e ea lie s udy. Acco ding o hese c i e ia, only one SNP,
s1555839, p e iously associa ed wi h apid au oma ized naming in
he Genes, Reading, and Dyslexia (GRaD) coho (7) yielded a sig-
nifican signal in he emainde o he GenLang coho s, showing
associa ion wi h spelling (P=3.33 ×10
4
). This SNP is one o
fi e SNPs ha eached he h eshold o genome-wide significance
in he o iginal GWAS o he GRaD coho .
Some 20 genes ha e been desc ibed in he li e a u e as candi-
da e genes o eading-/language- ela ed ai s and diso de s
based on a ange o mapping app oaches and ha e been he
ocus o much o he p io published esea ch in his a ea (4).
Da ase S11 gi es gene-based P alues om ou GenLang
−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
**
*
**
** **
**
***
**
***
***
***
***
***
***
***
***
***
***
*** ***
***
***
***
Pe o mance IQ
Educa ional
a ainmen
Full-scale IQ
GWAS by subs a ion
EA minus ull-scale IQ
Componen 1
O e all school pe o mance
Componen 2
Language be e han ma hs
Componen 3
O al exam be e han w i en
Componen 4
Danish be e han English
Wo d
eading
Nonwo d
eading
Spelling
Phoneme
awa eness
Nonwo d
epe i ion
*
**
*****
***
***
***
***
***
***
***
***
***
**
***
***
***
***
**
Wo d
eading
Nonwo d
eading
Spelling
Phoneme
awa eness
Nonwo d
epe i ion
A
B
1
.12 (.07) .06 (.12)
1
.001 (.08)
1
.001 (.06)
1
.21 (.04)
1
.94 (.05)
.97 (.06)
1.01 (.05)
.71 (.13)
1
uWR
Phoneme
awa enessg
uPA
Fac o 1g
1
Fac o 2g
Full-scale
IQg
uIQ
Nonwo d
eadingg
uNREAD uSP
Spellingg
.35 (.12)
.28 (.10)
.19 (.20)
1
.31 (.06)
1
Pe o mance
IQg
uPIQ
Educa ional
a ainmen g
uEA
.52 (.27)
1
Nonwo d
epe i iong
uNREP
1
Fac o 3g
.60 (.05)
.69 (.08)
.58 (.13)
.51 (.08)
.76 (.08)
.60 (.05)
.35 (.14)
.90 (.08)
Wo d
eadingg
Fig. 1. Reading- and language- ela ed ai s
ha e a sha ed gene ic a chi ec u e ha is
la gely independen o pe o mance IQ. (A)
Gene ic co ela ions ( g) among he eading-
and language- ela ed ai s es ima ed wi h
LDSC. Es ima es a e capped a one. Full LDSC
esul s a e epo ed in Da ase S4. In addi ion,
gene ic co ela ions a e gi en be ween he
GenLang ai s and 1) pe o mance IQ (using
GenLang coho s only); 2) educa ional a ain-
men (EA; n=766,345) and ull-scale IQ (n=
257,828) (18); 3) noncogni i e abili ies in ol ed
in EA, esul ing om a ecen GWAS by sub-
ac ion s udy (n=510,795) (19); and 4) com-
ponen s associa ed wi h dis inc pe o mance
domains iden ified used a decomposi ion
analysis o Danish school g ades (n=30,982)
(20). Full esul s can be ound in Da ase S8.
*Significan gene ic co ela ion a e co ec-
ion o 18.28 independen compa isons (P<
2.74 ×10
3
); **P<2.74 ×10
4
; ***P<2.74 ×
10
5
.(B) Th ee- ac o model fi ed o he Gen-
Lang summa y s a is ics o wo d eading,
nonwo d eading, spelling, phoneme awa e-
ness, nonwo d epe i ion, and pe o mance
IQ and o published GWAS summa y s a is ics
o ull-scale IQ and EA (18) using Genomic-
SEM (21). Black and g ay pa hs ep esen ac o
loadings wi h P<0.05 and P>0.05, espec-
i ely. S anda dized ac o loadings a e shown,
wi h SE in pa en heses. The subsc ip g ep e-
sen s he gene ic a iables; he u a iables ep-
esen he esidual gene ic a iance no
explained by he models. Uns anda dized
esul s and model fi indices a e epo ed in
Da ase S9.
4o 12 h ps://doi.o g/10.1073/pnas.2202764119 pnas.o g
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.

GWAS me a-analysis, calcula ed by Mul i-ma ke Analysis o
GenoMic Anno a ion (MAGMA) (22), o each o hese genes.
Va ia ion in one, namely DCDC2, showed associa ion wi h
nonwo d eading ha passed he significance h eshold o mul-
iple es ing o 20 genes and 2.15 independen GenLang ai s
(P<0.0012). DCDC2 was o iginally iden ified in a linkage
egion o dyslexia suscep ibili y, and SNPs in and nea his
gene we e subsequen ly associa ed wi h dyslexia in candida e
gene s udies (23), al hough some in es iga ions, including a
me a-analysis o se en s udies, ailed o suppo his (24). No
single candida e SNP highligh ed in p io s udies o DCDC2,
no inanyo he candida egene,wassignifican ly associa ed wi h
any ai s in he GWAS me a-analysis esul s a e co ec ion o
es ing o 54 SNPs and 2.15 independen GenLang ai s (P<4.
31 ×10
4
)(Da ase S11). These null findings highligh he
impo ance o la ge-scale s udies o obus iden ifica ion o com-
mon DNA a ia ion associa ed wi h eading-/language- ela ed
ai s and diso de s and wa an a ee alua ion o he con ibu-
ions o candida e genes ha a e p ominen om p io li e a u e.
A Mul i a ia e GWAS Analysis o GenLang T ai s Maximizes
SNP He i abili y. To imp o e he powe o ou GWAS me a-
analysis o ollow-up analyses, we ook ad an age o he high
gene ic co ela ions be ween he ai s by pe o ming a mul i a i-
a e GWAS analysis o wo d eading, nonwo d eading, spelling,
and phoneme awa eness wi h he Mul i T ai Analysis o GWAS
(MTAG) me hod (25). This app oach imp o es he e ec es i-
ma es o uni a ia e GWAS esul s pe SNP by inco po a ing
in o ma ion om he o he gene ically co ela ed ai s. Al hough
MTAG gene a es ou pu o each p ima y inpu ai , hese we e
ex emely simila as a consequence o he pa icula ly high gene ic
co ela ions be ween he ai s. Hence, he mul i a ia e esul s o
wo d eading we e used o all ollow-up analyses because he
uni a ia e wo d eading GWAS me a-analysis has he la ges sam-
plesize(27,180 o heEu opeanances yanalysiscompa ed
wi h 12,411 o 17,278 o he o he h ee ai s). Al hough no
indi idual SNP eached genome-wide significance in his mul i-
a ia e analysis (SI Appendix,Fig.S4), he app oach imp o ed
powe o ollow-up analyses in h ee ways: 1) by inco po a ing
all a ailable da a om he di e en ai s wi hou inc easing mul-
iple es ing bu den; 2) by educing e o a iance and he eby,
inc easing he p opo ion o pheno ypic a iabili y cap u ed by
gene ic a ia ions, wi h SNP-based he i abili y ha was highe
han any o he uni a ia e es ima es wi h a smalle SE (0.29, SE
=0.02);and3)becausejoin analysiso hedi e en ai smaxi-
mized he e ec i e sample size o he da ase . Indeed, MTAG
es ima ed he GWAS equi alen sample size o he mul i a ia e
esul s as 41,783 compa ed wi h 27,180 o he co esponding
uni a ia e esul s. These mul i a ia e GenLang GWAS esul s
we e used o ollow-up analyses u ilizing no only gene ic co e-
la ion app oaches, bu also o he genome-based me hods, such as
gene p ope y analysis and he i abili y pa i ioning, u he
explained below.
Assessing Links o Gene ics o Neu oana omical Va ia ion in
Reading-/Language-Rela ed Ci cui y. The neu obiological ci -
cui y in ol ed in spoken and w i en language has been ex en-
si ely in es iga ed in p io li e a u e, om pionee ing pos mo em
s udies o b ain lesions h ough o he nonin asi e s uc u al and
unc ional neu oimaging esea ch ha is now s anda d o he
field (26, 27). No ably, mul iple s udies ha e iden ified ela ion-
ships be ween measu es o neu oana omical ea u es and pe o -
mance on eading-/language- ela ed asks h ough analyses o
de elopmen al changes, o indi idual di e ences, and o ele an
diso de s (28, 29). A he same ime, a g owing numbe o la ge-
scale GWAS e o s ha e epo ed obus ela ionships be ween
common DNA a ia ion and indi idual di e ences in mul iple
aspec s o human neu oana omy, including b ain olumes, su -
ace a ea and hickness o di e en co ical egions, and whi e
ma e mic os uc u e, among o he s (30, 31). The a ailabili y o
GWAS da a om beha io al/cogni i e pheno ypes and om
magne ic esonance imaging (MRI)-based measu es o neu o-
ana omy makes i possible o de e mine gene ic o e laps be ween
b ains and beha io , e en when he s udy coho s a e inde-
penden (31, 32). He e, we applied his s a egy o assess gene ic
ela ionships be ween eading-/language- ela ed measu es and neu-
oana omical a ia ion, as ollows. Fi s , we pe o med a li e a u e
e iew o selec s uc u al neu oimaging ai s ha 1) encompass
b ain egions and whi e ma e ac s wi h known links o aspec s
o eading and language and 2) ha e been in es iga ed wi h
GWASin hela geUKBiobank esou ce(Ma e ials and
Me hods). This yielded 58 s uc u al neu oimaging ai s, includ-
ing su ace-based mo phome y (su ace a ea and hickness) (SI
Appendix,Fig.S6) pheno ypes and di usion enso imaging
(mean and weigh ed mean ac ional aniso opy) (SI Appendix,
Fig. S7) esul s. Second, as many o hese b ain-based pheno ypes
we e significan ly co ela ed wi h each o he , gene ic co ela ions
among hei summa y s a is ics we e used o calcula e he numbe
o independen ai s o mul iple es ing co ec ion. Thi d, we
pe o med gene ic co ela ion analyses be ween he mul i a ia e
GenLang GWAS esul s and he 58 neu oimaging ai s o iden-
i y gene ic o e laps. One neu oimaging ai showed significan
gene ic co ela ion wi h he mul i a ia e GenLang esul s (P<
0.05/24.85 independen ai s =2.01 ×10
3
): he su ace a ea o
he banks o he supe io empo al sulcus (STS) o he le hemi-
sphe e ( g =0.21, SE =0.06) (Fig. 2 and Da ase S12). This
finding sugges s he exis ence o sha ed gene ic ac o s ha con-
ibu e bo h o le STS su ace a ea and o eading-/language-
ela ed skills, albei wi hou iden i ying which SNPs unde lie he
ela ionship. Func ional MRI s udies ha e linked his egion o
di e en aspec s o w i en and spoken language p ocessing
(33–36). To in es iga e u he whe he he fi e di e en eading-
and language- ela ed ai s show simila o di e se gene ic co ela-
ions o banks o he le STS, we wen on o specifically assess
he esul s om heo iginaluni a ia eGenLangGWAS
me a-analyses, finding consis en gene ic co ela ions ( ange =
0.18 o 0.23) (Da ase S12).
Gene ic Co ela ion wi h T ai s om he UK Biobank and
B ain-Rela ed T ai s om he LD Hub. Nex , we assessed
gene ic co ela ions o ou mul i a ia e GWAS esul s wi h 20
cogni i e, educa ion, neu ological, psychia ic, and sleeping-
ela ed ai s and 515 addi ional UK Biobank ai s using LD
Hub. To u he in es iga e o e laps wi h and di e ences om
IQ, gene ic co ela ions be ween hese 535 ai s and he pub-
lished GWAS summa y s a is ics o ull-scale IQ (18) (n=
257,828) we e ob ained as well. A o al o 143 ai s showed
significan gene ic co ela ions wi h he mul i a ia e GenLang
GWAS esul s a e co ec ion o mul iple es ing [P<0.05/
(535 ×2) =4.67 ×10
5
](Da ase S13), while 245 ai s we e
gene ically co ela ed wi h ull-scale IQ; 135 ai s showed sig-
nifican co ela ions wi h bo h ou mul i a ia e GWAS and
ull-scale IQ. T ai s wi h s ong gene ic co ela ions we e ela ed
o educa ion, eyesigh , ch ono ype, well-being, li es yle, physical
heal h and exe cise, and socioeconomic s a us. Rep esen a i e
ai s a e plo ed in Fig. 3. Cogni i e and educa ion- ela ed ai s
showed highe gene ic co ela ions wi h ull-scale IQ han wi h
he mul i a ia e GenLang GWAS da a. Se e al psychia ic and
PNAS 2022 Vol. 119 No. 35 e2202764119 h ps://doi.o g/10.1073/pnas.2202764119 5o 12
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.
well-being ai s showed significan (nega i e) gene ic co ela-
ions wi h ull-scale IQ bu no wi h ou mul i a ia e GWAS:
o example, dep essi e symp oms; c oss-diso de suscep ibili y
( om he Psychia ic Genomics Conso ium GWAS); and ense,
hu , and ne ous eelings. In con as , se e al ai s ela ed o
physical heal h and li es yle showed la ge gene ic co ela ions
wi h he mul i a ia e GenLang GWAS esul s han wi h ull-
scale IQ, including body mass index, educed alcohol in ake as a
heal h p ecau ion, and usual walking pace.
Analysis o Genomic Anno a ions Rela ed o Human E olu ion.
W i ing and eading a e ela i ely ecen cul u al inno a ions,
bu mul iple lines o e idence indica e ha he ele an skills,
especially hose in ol ed in decoding as assessed he e, a e based
on ou capaci ies o spoken language, which eme ged h ough
biological e olu ion along he lineage ha led o humans
(37, 38). As no ed abo e, ou mul i a ia e GWAS cap u ed
common DNA a ia ion accoun ing o a subs an ial p opo ion
o in e indi idual a iabili y in eading-/language- ela ed ai s
wi hin ou s udy popula ions. We wen on o es in agg ega e
whe he he e we e o e laps be ween he genomic egions d i -
ing his associa ion signal and egions in ol ed in aspec s o
human e olu ion o e a ange o imescales. To do so, we used
LDSC he i abili y pa i ioning (39), a me hod ha uses GWAS
esul s o in es iga e whe he common DNA a ian s in a ce ain
se o genomic egions, named an anno a ion, explain a la ge p o-
po iono heSNP-basedhe i abili yo he ai hanisexpec ed
based on he size o ha anno a ion. Building on p io wo k on
e olu ion o human b ain s uc u e (40), we s udied fi e anno a-
ions eflec ing di e en aspec s o human e olu ion spanning
pe iods om 30 Mya o 50,000 y ago (SI Appendix,Fig.S8). The
es ed anno a ions included human gained enhance s ac i e in
e al and adul b ain issue, ancien selec i e sweep egions, Nean-
de hal in og essed a ian s, and a chaic in og ession dese s
(de ails a e in SI Appendix,Ex ended Me hods). The la e wo
anno a ions ela e o admix u e e en s be ween Homo sapiens and
Neande hal popula ions ha ook place when he di e en homi-
nins encoun e ed each o he ou side A ica some 50,000 o
60,000 y ago, wi h he consequen gene flow lea ing emnan s
(in og essed agmen s) ha can be de ec ed in he genomes o
li ing humans. A chaic dese s a e long s e ches in he human
genome ha , despi e hese admix u e e en s, a e significan ly
deple ed o Neande hal alleles in li ing humans, possibly due o
c i ical unc ions o he gene ic loci in H. sapiens and in ole ance
o gene flow (41). We obse ed significan ly en iched he i abili y
o a chaic dese s, which was obus o mul iple es ing co ec-
ion o analysis o fi e anno a ions (P<0.01) (Da ase S14).
Thus, ou esul s sugges ha common DNA a ia ion in
a chaic dese s makes a la ge con ibu ion o indi idual di e -
ences in eading-/language- ela ed ai s wi hin p esen -day pop-
ula ions han expec ed by chance.
Func ional En ichmen Using He i abili y Pa i ioning and
MAGMA Gene P ope y Analysis. We nex in es iga ed whe he
egions o he genome wi h issue-specific unc ions a e in ol ed
in eading-/language- ela ed ai s, again using LDSC he i abili y
pa i ioning. We used anno a ions eflec ing ch oma in signa u es
om a b oad ange o issues, as mos a ian s iden ified in
GWAS a e loca ed ou side coding egions and a e o en ound
en iched in issue-specific unc ional egions o he genome, such
as p omo e s, enhance s, and egions wi h open ch oma in. A e
co ec ion o es ing o 489 anno a ions (P<1.02 ×10
4
), h ee
anno a ions showed significan he i abili y en ichmen : his one-3
lysine-4 monome hyla ion (H3K4me1) in wo e al b ain samples
and he adul b ain ge minal ma ix (SI Appendix, Fig. S9 and
Da ase S14). H3K4me1 is conside ed a ma ke o enhance
egions. These esul s indica e ha SNPs associa ed wi h eading-/
language- ela ed ai s a e o e ep esen ed in e al b ain enhance s.
Nex , we used MAGMA gene p ope y analysis (22) o s udy
whe he he mul i a ia e GenLang GWAS esul s we e en iched
in a specific issue o b ain cell ype using issue-specificandcell
ype–specific gene exp ession da a in Func ional Mapping and
Anno a ion (FUMA) (42, 43). As MAGMA co ec s o a e age
exp ession, each compa ison can only answe he ques ion o
whe he he issue o cell ype is mo e ela ed o he mul i a ia e
GWAS esul s han he a e age o he issues o cell ypes in he
da ase . A e co ec ion o 83 issues (P<6.02 ×10
4
), no ela-
ion was ound wi h issue-specific gene exp ession pa e ns o
adul issues om he Geno ype-Tissue Exp ession (GTEx) p ojec
and b ain issues o a specific(de elopmen al) ime omB ain-
span (Da ase S15 and SI Appendix,Fig.S10). In he cell
ype–specific gene exp ession analysis, h ee single-cell RNA-
sequencing da ase s o emb yonic, e al, and adul b ain issue
we e used. A e co ec ion o 142 cell ypes (P<3.52 ×10
4
),
he mul i a ia e GWAS esul s we e significan ly associa ed wi h
one o he ma u e neu ons om he e al da ase : ed nucleus neu-
ons (be a =0.24, SE =0.07) (Da ase S15 and SI Appendix,Fig.
S11). This obse a ion could eflec an associa ion wi h his spe-
cific nucleus o wi h he highe ma u i y o he neu ons compa ed
wi h he o he cell ypes in he e al da ase . The ed nucleus is a
la ge s uc u e in he en al midb ain ha is pa o he oli oce e-
bella and ce ebello– halamo–co ical sys ems. I plays oles in
locomo ion and nonmo o beha io in a ious animals, and in
humans, i migh also play a ole in highe co ical unc ions (44).
Discussion
We pe o med GWAS me a-analysis o fi e quan i a i e
eading- and language- ela ed ai s (wo d eading, nonwo d
eading, spelling, phoneme awa eness, and nonwo d epe i ion)
in sample sizes (up o ∼34,000 pa icipan s) ha a e subs an-
ially la ge han p e ious gene ic analyses o eading and/o lan-
guage skills assessed wi h neu opsychological ools (p io GWAS
e o s wi h maximal sizes o n=1,331 o 10,819) (5–7, 45–47).
We iden ified genome-wide significan associa ion o wo d eading
( s11208009 a ch omosome 1, P=1.10 ×10
8
), highligh ing
DOCK7,ATG4C,ANGPTL3,andUSP1 as po en ial candi-
da es o in ol emen in his ai . O he SNPs om he locus
ha e been associa ed wi h iglyce ide and choles e ol le els (48)
bu may ep esen an independen associa ion signal. Robus
SNP-based he i abili ies we e obse ed, anging om 0.13 o
*
0.25
0.2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.20
-0.25
g
Fig. 2. The mul i a ia e GenLang GWAS esul s show significan gene ic
co ela ion wi h he co ical su ace a ea a ound he le STS. Gene ic co e-
la ions ( g) we e es ima ed wi h LDSC. Included ai s a e 58 s uc u al
b ain imaging ai s om he UK Biobank selec ed based on known links o
egions and ci cui s wi h language p ocessing. The esul s o he 22 co ical
su ace a eas a e shown; g ay a eas we e no included in he analysis.
Full esul s can be ound in Da ase S12 and SI Appendix, Figs. S6 and S7.
*Significan gene ic co ela ion a e co ec ing o 24.85 independen b ain
imaging ai s (P<2.01 ×10
3
).
6o 12 h ps://doi.o g/10.1073/pnas.2202764119 pnas.o g
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.
nonwo d epe i ion o 0.26 o nonwo d eading. These SNP-
based he i abili ies a e simila in magni ude o hose o he
ela ed ai dyslexia (es ima es ange om 0.15 o 0.25 on a
liabili y scale) (49, 50); psychia ic ai s, such as A en ion Defi-
ci Hype ac i i y Diso de symp oms, in adul s (0.22) (51); and
b ain imaging ai s, such as co ical su ace a ea ( ange om 0.12
o 0.33 o di e en egions) (52) and co ical hickness ( ange
om 0.08 o 0.26) (52). They a e la ge han ha o psychia ic
ai s, such as majo dep ession (0.08) (53) and alcohol depen-
dence (0.09) (54). So, despi e highligh ing he need o la ge
sample sizes o iden i y loci ha indi idually exceed genome-wide
significance, ou GenLang GWAS al eady allowed o mul iple
in o ma i e ollow-up analyses based on he ull da ase o com-
mon a ian s ac oss he genome, yielding findings ha connec o
neu oana omical a ia ion, human e olu iona y his o y, and o he
aspec s o he biology o spoken and w i en language.
Ou wo k shows o e lapping con ibu ions o common
gene ic a ian s o indi idual di e ences in eading-/language-
ela ed and cogni i e ai s. Such o e laps a e e iden bo h
om gene ic co ela ion analyses and om he h ee- ac o
s uc u al equa ion model, in which one ac o explains mos o
he a ia ion in wo d eading, nonwo d eading, spelling, and
phoneme awa eness. This is in line wi h he widesp ead pleio -
opy ound be ween many aspec s o cogni i e unc ioning,
including language, eading, ma hema ics, and gene al cogni-
i e abili y (6, 55). We no e ha summa y s a is ics om he
cu en in es iga ion ha e also been used o in es iga e gene ic
o e laps wi h sel - epo o dyslexia diagnosis in an independen
GWAS by 23andMe (∼52,000 cases), yielding subs an ial nega-
i e gene ic co ela ions be ween GenLang quan i a i e ai s
and dyslexia s a us (e.g., 0.71 o wo d eading, 0.75 o
spelling) as epo ed by Dous e al. (50). Ye , nonwo d epe i-
ion, IQ, and educa ional a ainmen ha e, a leas in pa , di -
e en gene ic ounda ions, as eflec ed in he esidual gene ic
a ia ion con ibu ing o hese ai s ha is no cap u ed by he
model. These findings a e consis en wi h mul iple beha io al
s udies showing a dis inc ion be ween nonwo d epe i ion and
o he eading- and language- ela ed ai s (56). They a e also
in line wi h ecen s uc u al equa ion modeling o gene ic
ai in e ela edness o di e en eading- and language- ela ed
Cogni i e
Educa ion
Eyesigh
Ch ono ype
Li es yle (UKBB)
Wellbeing (UKBB)
Psychia ic
Pain (UKBB)
Physical heal h and exe cise (UKBB)
SES (UKBB)
Noisy wo kplace
Job in ol es walking o s anding
Job in ol es hea y physical wo k
Job in ol es shi wo k
Financial si ua ion dissa is ac ion
Townsend dep i a ion index a ec ui men
Age a las li e bi h
Age a i s li e bi h
Paid empoymen o sel −employed
Unpaid o olun a y job
Usual walking pace
F equency mode a e physical ac i i y
F equency walking a leas 10 minu es
F equency igo ous physical ac i i y
O e all heal h c i ique
Den u es
Dias olic blood p essu e
Sys olic blood p essu e
Body a pe cen age
BMI
Hip ci cum e ence
Fo ced expi a o y olume
Fo ced i al capaci y
Peak expi a o y low
S anding heigh
Pa ace amol use
Pain all o e he body
Knee pain
Hip pain
Neck o shoulde pain
Back pain
No pain
PGC c oss-diso de (me a-analysis 2013)
Dep essi e symp oms (me a-analysis 2016)
A
u ism spec um diso de (me a-analysis PGC)
Ne ous eelings
Sensi i i y / hu eelings
Tense / highly s ung
Loneliness o isola ion
Fed−up eelings
Mood swings
F equency o unen husiasm
F equency o enseness / es lessness
F equency o dep essed mood
F iendships dissa is ac ion
Time spen wa ching TV
Time spen d i ing
Time spen using compu e
D i e oo as
S opped smoking
S opped smoking as heal h p ecau ion
Cu en smoking
Ne e d ank alcohol
Reduced alcohol in ake as heal h p ecau ion
Alcohol in ake
Bee and cide in ake
Red wine in ake
Whi e wine in ake
Mo ning/e ening pe son (UKBB)
Glasses o con ac lenses (UKBB)
Glasses o sho −sigh edness (UKBB)
Glasses o as igma ism (UKBB)
No quali ica ions (UKBB)
CSE quali ica ion (UKBB)
NVQ o HND o HNC quali ica ion (UKBB)
Nu sing o eaching quali ica ion (UKBB)
O le els/GCSEs quali ica ion (UKBB)
A le el quali ica ion (UKBB)
College comple ion (me a-analysis 2013)
College o Uni e si y deg ee (UKBB)
Age comple ed ull ime educa ion (UKBB)
Yea s o schooling (me a-analysis 2016)
Numbe o inco ec ma ches (UKBB)
P ospec i e memo y es di icul y (UKBB)
Childhood IQ (me a-analysis 2014)
Fluid in elligence (UKBB)
−1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8 1.0
Gene ic co ela ion ( g)
Mul i a ia e GenLang Full-scale IQ
Fig. 3. Gene ic co ela ion esul s o he mul-
i a ia e GenLang GWAS analysis wi h com-
pa isons wi h hose o he la ges published
GWAS o ull-scale IQ in LD Hub. Summa y s a-
is ics o ull-scale IQ (n=257,828) we e
ob ained om he Social Science Gene ic
Associa ion Conso ium (18). Gene ic co ela-
ions be ween he mul i a ia e GenLang
esul s (blue–g een), ull-scale IQ (pu ple), and
ai s in LD Hub e eal an o e lap wi h cogni-
i e ai s, educa ion, eyesigh , ch ono ype,
li es yle, well-being, psychia ic diso de s,
pain, physical heal h and exe cise, and socio-
economic s a us. A subse o ep esen a i e
ai s is shown; 143 ai s showed significan
associa ions wi h he mul i a ia e GenLang
esul s, and 245 ai s showed significan co -
ela ions wi h ull-scale IQ, o which 135 ai s
o e lap a e co ec ion o mul iple es ing
o 535 ×2 ai s (P<4.67 ×10
5
). Significan
co ela ions a e shown in da k colo s; nonsig-
nifican co ela ions a e in ligh colo s. Full
esul s can be ound in Da ase S13. UKBB: UK
Biobank, GCSE: Gene al Ce ifica e o Second-
a y Educa ion, NVQ: Na ional Voca ional Quali-
fica ion, HND: Highe Na ional Diploma, HNC:
Highe Na ional Ce ifica e, CSE: Ce ifica e o
Seconda y Educa ion, PGC: Psychia ic Geno-
mics Conso ium, BMI: body mass index, SES:
socioeconomic s a us. Gene ic co ela ion ( g)
is p esen ed as a do , and e o ba s indica e
he SE.
PNAS 2022 Vol. 119 No. 35 e2202764119 h ps://doi.o g/10.1073/pnas.2202764119 7o 12
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.
measu es in he A on Longi udinal S udy o Pa en s and Chil-
d en (ALSPAC) coho , which demons a ed a sha ed gene ic
ac o accoun ing almos ully o he gene ic a iance in
li e acy- ela ed pheno ypes bu o only 53% o ha in non-
wo d epe i ion (57). The 23andMe GWAS on sel - epo ed
dyslexia u he shows he exis ence o e ec s on eading abili y
ha a e independen o IQ; o 42 genome-wide significan
loci in ha s udy, less han hal had p e iously been associa ed
wi h cogni i e abili y o educa ional a ainmen in p io high-
powe ed in es iga ions (50). O e all, ou wo k hus enhances
unde s anding o no only he o e laps bu also, he dis inc ions
be ween di e en eading-/language- ela ed ai s and mo e gen-
e al cogni i e abili ies.
Fu he e idence o di e ences in ai e iology be ween he
fi e eading-/language- ela ed ai s was obse ed in ou gene ic
co ela ion analysis wi h GWAS esul s o componen s (iden ified
ia decomposi ion analysis) om school g ades in he Danish
iPSYCH coho (20) and wi h esul s o a GWAS by sub ac ion
analysis o educa ional a ainmen and cogni i e pe o mance
(19). Highe sco es on phoneme awa eness and spelling appea o
be gene ically co ela ed wi h be e pe o mance in w i en han
in o al examina ions. This may eflec he g ea e impo ance o
phoneme awa eness and spelling o p oficien w i ing han o
o al language. In e es ingly, wo d and nonwo d eading skills
we e associa ed wi h be e pe o mance in language han
ma hema ics bu no wi h be e pe o mance in w i en han
o al language. Such findings may o e u he p oo ha key
eading skills, especially hose in ol ed in decoding as assessed
in ou GWAS me a-analyses, o igina e om o al language
skills (58). “Componen 4,”co esponding o ela i ely be e
pe o mance in Danish ( he na i e language o he pa icipan s in
ha s udy) as compa ed wi h pe o mance in English, showed
nega i e gene ic co ela ions wi h ou GenLang nonwo d
epe i ion measu e, possibly eflec ing he pa icula impo ance
o e bal sho - e m memo y in second-language lea ning (59,
60). The esul s o he GWAS by sub ac ion, p e iously
p oposed o ep esen so-called “noncogni i e”abili ies ela ed o
educa ional a ainmen , such as mo i a ion, cu iosi y, and
pe sis ence, we e gene ically co ela ed wi h wo d and nonwo d
eading as well as nonwo d epe i ion in GenLang. O no e,
gene ic co ela ion analyses may be influenced by gene ic nu u e,
he p ocess whe eby DNA a ian s o he pa en s a ec
pheno ypic ou comes in hei child en (61). Gene ic a ian s
ela ing o he socioeconomic s a us o he amily may, o
example, be in ol ed, as was ecen ly ound o cogni i e ai s
(62). Fu u e in es iga ions ha include in o ma ion abou
non ansmi ed alleles (63) and/o da a om siblings (62) may
help o disen angle pleio opy om gene ic nu u e.
Humanabili ies op ocessspokenandw i enlanguagedepend
on an a ay o dis ibu ed b ain ci cui s (28, 29, 64–66). We pe -
o med gene ic co ela ion analyses o ou mul i a ia e GenLang
GWAS wi h summa y s a is ics om 58 MRI-based neu oana om-
ical pheno ypes chosen because hey conce ned b ain a eas and/o
ac s wi h known links o language p ocessing (28, 29, 65, 66).
We iden ified a significan gene ic co ela ion wi h co ical su ace
a ea o he banks o he STS on he le hemisphe e. The STS is a
egion whe e he p ocessing o spoken and w i en language con-
e ges, in be ween modali y-specific p ep ocessing and language
comp ehension (33–36). A b oad ange o language- ela ed unc-
ions has been p e iously linked wi h he le STS h ough (me a-
analyses o ) unc ional MRI and posi on emission omog aphy
s udies, including hose essen ial o he eading- and language-
ela ed ai s included in he GWAS me a-analyses: sublexical p oc-
essing o speech (67, 68) and ep esen a ion o phonological wo d
o ms (69). The impo ance o his b ain a ea o eading- ela ed
ai s is also e iden om a me a-analysis o s uc u al MRI s udies
ha ound lowe g ay ma e olume in he le STS ela ed o
eading disabili y and poo eading comp ehension (70). Thus,
findings om he gene ic co ela ion analysis a e consis en wi h
he ole o he STS as a hub whe e he p ocessing o di e en lan-
guage modali ies ge s in eg a ed as well as he la e aliza ion o
such unc ions. No e, howe e , ha while ou wo k suppo s he
exis ence o sha ed gene ic ac o s influencing bo h le STS su -
ace a ea and psychome ic measu es o eading-/language- ela ed
skills, i does no in o m abou po en ial causal ela ionships o
di ec ion o e ec s, no does i iden i y which pa icula molecula
mechanisms may be in ol ed.
Capaci ies o acqui ing spoken and w i en language appea
unique o ou species, building on unde lying skills ha eme ged on
he lineage leading o mode n humans, bu e olu iona y accoun s
emain subjec o conside able deba e (37, 38). To explo e whe he
GWAS da a could gi e insigh s in his a ea, we used he i abili y pa -
i ioning o analyze fi e anno a ions ep esen ing di e en ime
ames and aspec s o human e olu ion. A chaic in og ession
dese s, defined as genomic egions ha a e significan ly deple ed o
Neande hal ances y, we e en iched o gene ic a ian s showing
associa ions in ou mul i a ia e GenLang GWAS. Such egions a e
hough o co espond o genomic loci ha we e in ole an o he
gene flow om Neande hal popula ions in o H. sapiens,which
ook place h ough admix u e e en s a ound 50,000 o 60,000 y
ago (41). These loci a e en iched o conse ed and unc ional geno-
mic elemen s: p omo e s and egions conse ed in p ima es (71)
andenhance sac i einmany issuesaswellas hosespecific o
exp ession in e al b ain and muscle (72). O all egula o y egions
in a chaic dese s, b ain enhance s show signs o he mos s ingen
pu i ying selec ion agains in og essed Neande hal a ia ion. This
e idence o conse a ion and selec ion indica es ha a chaic dese s
ma k pa s o he genome whe e a ia ion has a high p obabili y o
dele e ious consequence (72). En iched he i abili y o eading-/lan-
guage- ela ed ai s in a chaic dese s seems b oadly consis en wi h
di e ences be ween Neande hals and H. sapiens in e olu iona y a-
jec o ies o language eme gence. Howe e , analyses o his kind a e
only indi ec ly in o ma i e and canno pinpoin any specific ime
ame o human e olu ion du ing which gene ic a ian s associa ed
wi h eading- and language- ela ed ai s we e in oduced. In es iga-
ions ha u he in eg a e da a on eading-/language- ela ed gene ic
signals and e olu iona y anno a ions o hominin genomes would be
wa an ed o shed u he ligh on hese complex ques ions.
Rega ding unc ional implica ions o ou findings, he i abili y
pa i ioning o he mul i a ia e GenLang GWAS esul s iden ified
en ichmen in enhance egions p esen in e al b ain issue and
he adul ge minal ma ix. The enhance egions o he ge minal
ma ix a e highly simila o hose o e al b ain issues and no he
o he adul b ain issues we s udied, likely eflec ing he neu al
s em cell popula ion p esen in ha issue (73). In hese analyses,
he e was no specific associa ion wi h one pa icula b ain cell
ype. Howe e , ollow-up wo k wi h MAGMA using single-cell
RNA-sequencing da a om e al, emb yonic, and adul b ain
unco e ed a significan associa ion wi h e al neu ons om he ed
nucleus, which may ela e o he mo e adul s a e o hese neu ons
compa ed wi h he o he cell ypes in he e al da ase (74). The
MAGMA analysis could no be used o es o eplica ion o he
associa ion wi h ( e al) b ain issue o he LDSC he i abili y pa i-
ioning, as esul s in his case a e co ec ed o he associa ion wi h
he a e age exp ession o he da ase (22, 42), and he da ase s
wi h e al da a only included b ain samples.
Ou findings p omp a majo ee alua ion o p io li e a u e
on gene ic associa ions wi h eading and language ai s, especially
8o 12 h ps://doi.o g/10.1073/pnas.2202764119 pnas.o g
Downloaded om h ps://www.pnas.o g by BIN 8102 KIRJASTO KAUSI JUL on Augus 25, 2022 om IP add ess 130.234.230.116.