Large-scale GWAS identifies multiple loci for hand grip strength providing biological insights into muscular fitness
Full text
ARTICLE
Recei ed 13 Ma 2017 |Accep ed 22 May 2017 |Published 12 Jul 2017
La ge-scale GWAS iden ifies mul iple loci o hand
g ip s eng h p o iding biological insigh s in o
muscula fi ness
Sa a M. Willems e al.#
Hand g ip s eng h is a widely used p oxy o muscula fi ness, a ma ke o ail y, and
p edic o o a ange o mo bidi ies and all-cause mo ali y. To in es iga e he gene ic de e -
minan s o a ia ion in g ip s eng h, we pe o m a la ge-scale gene ic disco e y analysis in a
combined sample o 195,180 indi iduals and iden i y 16 loci associa ed wi h g ip s eng h
(Po5108) in combined analyses. A numbe o hese loci con ain genes implica ed in
s uc u e and unc ion o skele al muscle fib es (ACTG1), neu onal main enance and signal
ansduc ion (PEX14, TGFA, SYT1), o monogenic synd omes wi h in ol emen o psycho-
mo o impai men (PEX14, LRPPRC and KANSL1). Mendelian andomiza ion analyses a e
consis en wi h a causal e ec o highe gene ically p edic ed g ip s eng h on lowe ac u e
isk. In conclusion, ou findings p o ide new biological insigh in o he mechanis ic unde -
pinnings o g ip s eng h and he causal ole o muscula s eng h in age- ela ed mo bidi ies
and mo ali y.
Co espondence and eques s o ma e ials should be add essed o R.S. (email: obe .sco @m c-epid.cam.ac.uk) o o N.W.
(email: nick.wa eham@m c-epid.cam.ac.uk).
#A ull lis o au ho s and hei a filia ions appea s a he end o he pape .
DOI: 10.1038/ncomms16015 OPEN
NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions 1
Muscle s eng h, measu ed by isome ic hand g ip
s eng h, is an accessible and widely used p oxy o
muscula fi ness. Lowe g ip s eng h is associa ed wi h
impai ed quali y o li e in olde adul s, and is an es ablished
ma ke o ail y, p edic ing physical decline and unc ional
limi a ion in daily li ing1–3. The alue o g ip s eng h as a
clinical p edic o o ac u e isk has been demons a ed in
di e en popula ions4,5, and highe g ip s eng h has been ound
o be p ognos ic o walking eco e y a e hip ac u e su ge y in
la e li e6. G ip s eng h has also been shown o p edic
ca dio ascula disease (CVD) and all-cause mo ali y o e
many yea s o ollow-up7–9. Whils i emains unclea whe he
hese p ospec i e associa ions wi h ac u e isk, CVD and
mo ali y a e causal—o eflec ea ly mani es a ion o
unde lying disease p ocesses— he ole o muscula s eng h as a
p edic o o unc ional capaci y highligh s he impo ance o
unde s anding i s ae iology.
G ip s eng h is highly he i able (h2¼30–65%)10–12. Whils
candida e gene app oaches ha e implica ed mul iple loci in his
pheno ype, including he mogenic and myogenic ac o s13,14,
he e emain ew obus ly eplica ed associa ions. Two genome-
wide associa ion s udies in up o 27,000 indi iduals ha e been
epo ed o da e15,16, yielding one in e genic genome-wide
significan associa ion16.
He e, in a combined sample size o 195,180 indi iduals,
including 142,035 indi iduals om he UK Biobank (UKB)
coho 17, we iden ified 16 genome-wide significan loci associa ed
wi h g ip s eng h. We also pe o med Mendelian andomiza ion
(MR) analyses, which showed no e idence o causali y in
he associa ions o g ip s eng h wi h CVD o all-cause mo ali y,
bu we e sugges i e o a causal e ec o muscula s eng h on
ac u e isk.
Resul s
Mul iple no el loci a e associa ed wi h g ip s eng h.Ins age
one analyses, we es ed he associa ion o 417 million a ian s
(mino allele equency (MAF)40.1%, impu a ion quali y 40.4),
in 142,035 whi e Eu opean indi iduals om UK Biobank
(Supplemen a y Table 1) wi h maximal g ip s eng h. Genome-wide
single-nucleo ide a ian (SNV) he i abili y was es ima ed a 23.9%
(SE 2.7%). Twen y-one loci showed genome-wide significan asso-
cia ions (Po5108) in s age one (Supplemen a y Fig. 1), and
we e subsequen ly ollowed up in s age wo analyses o up o 53,145
indi iduals om 8 addi ional s udies (Supplemen a y Table 1;
Supplemen a y No e) including he Coho s o Hea and Aging
Resea ch in Genomic Epidemiology (CHARGE) conso ium16.
Twel e loci we e independen ly eplica ed (di ec ional consis ency
wi h s age one, Po0.05) in s age wo coho s (Supplemen a y
Table 2A) and 16 loci con ained genome-wide significan
associa ions (Po5108) in combined analyses. E ec sizes on
g ip s eng h anged om 0.14 o 0.42kg pe allele unde an
addi i e model (Table 1; Supplemen a y Fig. 1; Supplemen a y
Table 2A and B). Gi en he disco dance in sample size be ween
s age one and wo analyses, and in he in e es s o maximizing
powe , we conside ed he e o be e idence o associa ion a any
locus eaching genome-wide significance in combined analyses, and
pu sued all 16 in downs eam analyses. Lead SNVs a he 16 g ip
s eng h-associa ed loci included common a ian s (MAFZ5%) in
o nea POLD3,TGFA,ERP27,HOXB3,GLIS1,PEX14,MGMT,
LRPPRC, SYT1, GBF1, KANSL1, SLC8A1, IGSF9B, ACTG1, alow-
equency a ian (MAF 3%) in DEC1, and a u he common
a ian alling wi hin he human leukocy e an igen (HLA) egion
(Table 1; Supplemen a y Fig. 2). App oxima e condi ional analyses
iden ified no addi ional signals a genome-wide significance a hese
16 loci a e condi ioning on hei espec i e lead SNVs. A wo loci,
we saw e idence o a depa u e om addi i i y (Po3.13 103
unde a dominance de ia ion model (see Me hods)); a he GBF1
locus, we saw e idence o a dominan e ec o he g ip s eng h-
aising A allele (P
domde
¼2.3 103;Supplemen a yFig.3A),and
a he SYT1 locus, we saw e idence o a ecessi e e ec o he g ip
s eng h- aising A allele (P
domde
¼3.0 103;Supplemen a y
Fig. 3B). No indi idual a ian s showed significan e ec
modifica ion by age o sex (Supplemen a y Table 2C and D). The
associa ion o he 16 SNV gene ic sco e (modelled as he sum
o he g ip s eng h-inc easing allele dosage a each SNV pe
indi idual) showed no in e ac ion wi h age (P
in e ac ion
¼0.30),
bu was s onge in men han in women (men: b¼0.20 kg pe
g ip s eng h-inc easing allele, P¼2.38 1048;women:
b¼0.13 kg pe g ip s eng h-inc easing allele, P¼3.61 10 43;
P
in e ac ion
¼1.56 105;Fig.1;Supplemen a yTable2CandD).
Age a ec ui men was independen o s eng h-inc easing allele
Table 1 | Associa ion o he six een loci eaching genome-wide significance in combined analyses.
S age one (UKB)wS age wo coho s Combined
sID Gene* All. EAF E ec zS.E. P- alue E ec zS.E. P- alue E ec zS.E. P- alue N
s958685 TGFA A/C 0.52 0.154 0.026 2.8 1090.164 0.04 3.8 1050.157 0.022 4.8 10 13 191,754
s72979233 POLD3 A/G 0.76 0.210 0.03 3.7 1012 0.112 0.041 5.8 10 30.175 0.024 5.0 1013 192,490
s11614333 ERP27 C/T 0.62 0.181 0.027 5.0 1011 0.117 0.04 3.5 1030.16 0.023 1.6 1012 195,154
s2288278 HOXB3 A/G 0.66 0.162 0.027 3.0 1090.147 0.04 2.8 10 40.157 0.023 3.8 1012 195,133
s4926611 GLIS1 C/T 0.64 0.173 0.027 1.3 1010 0.115 0.041 5.1 1030.156 0.023 4.8 10 12 192,964
s6687430 PEX14 G/A 0.46 0.15 0.026 7.6 1090.124 0.04 1.7 1030.142 0.022 5.6 10 11 195,176
s10186876 LRPPRC A/G 0.36 0.162 0.027 2.7 10 90.113 0.041 6.2 10 30.147 0.023 9.8 1011 192,490
s374532236 MGMT T/C 0.38 0.157 0.027 5.5 1090.121 0.042 4.2 1030.147 0.023 1.1 10 10 189,701
s10861798 SYT1 A/G 0.43 0.145 0.026 4.3 1080.159 0.047 7.4 1040.148 0.023 1.3 1010 189,160
s78325334 HLA T/C 0.84 0.228 0.038 2.4 10 90.113 0.05 0.024 0.186 0.03 9.6 10 10 193,127
s2273555 GBF1 A/G 0.61 0.153 0.027 9.1 1090.096 0.041 0.019 0.136 0.022 1.1 109191,754
s80103986 KANSL1 A/T 0.81 0.201 0.033 1.8 1090.098 0.052 0.059 0.171 0.028 1.2 10 9193,090
s2110927 SLC8A1 C/T 0.27 0.161 0.029 4.4 10 80.098 0.045 0.029 0.142 0.025 7.7 109192,490
s6565586 ACTG1 A/T 0.25 0.169 0.03 2.2 10 80.096 0.064 0.14 0.156 0.027 1.2 108187,072
s72762373 DEC1 A/G 0.03 0.424 0.078 4.9 10 80.359 0.255 0.16 0.418 0.074 1.8 10 8152,162
s34845616 IGSF9B A/G 0.25 0.168 0.03 1.7 10 80.07 0.049 0.15 0.141 0.025 2.7 108189,666
All, alleles (e ec /o he ); EAF, e ec allele equency; HLA, HLA egion; N, sample size; UKB, UK Biobank.
Resul s a e so ed by combined s age one þs age wo P- alue.
*Nea es gene o he lead SNP.
wS age one analyses include 142,035 pa icipan s.
zE ec es ima es a e in kg pe allele and co espond o he fi s allele shown.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015
2NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions
dosage a each o he 16 SNVs om combined analyses. Equally,
allele equency a each SNP was no p edic ed by age, sugges ing
ha he e is no selec ion o alleles by age a hese loci18.Wedidno
eplica e he p e iously- epo ed associa ion a s752045 wi h g ip
s eng h16 (bpe mino allele (95% confidence in e al (CI))¼0.01
(0.06, 0.08), P¼0.75).
A numbe o associa ed loci con ained genes wi h biologically
plausible oles in s eng h and neu omuscula fi ness, h ough
e ec s on he s uc u e and unc ion o skele al muscle (ACTG1),
exci a ion-con ac ion coupling (SLC8A1), e idence o neu o-
ophic oles (TGFA), o in ol emen in he egula ion o
neu o ansmission (SYT1). ACTG1 (Ac in, g1A) encodes a key
componen o he cos ame e—a p o ein complex localized o he
Z-disc o skele al muscle which physically e he s myofib ils o he
cell memb ane and ansmi s con ac ile o ce gene a ed a he
sa come e o he ex acellula ma ix ia he dys ophin glycop o-
ein complex (DGC)19,20. Monogenic loss o elemen s o he DGC
esul s in muscula dys ophies21, whils Ac g1 knockou mice
display o e muscle weakness, p og essi e myopa hy and
dec eased isome ic wi ch o ce22.SLC8A1 encodes a
ansmemb ane Naþ/Ca2þexchange which is i al o es o ing
Ca2þconcen a ion o p e-exci a ion le els in exci able cells.
Muscle-specific o e exp ession o SLC8A1 has been shown o
induce dys ophy-like skele al muscle pa hology23. Synap o agmin-
1, encoded by SYT1, is an in eg al synap ic memb ane p o ein
which egula es Ca2þ-dependen neu o ansmi e elease a he
p esynap ic e minal24, and is implica ed in de elopmen o
neu omuscula junc ion pa hology in oden models o spinal
muscula a ophy25.TGFA encodes ans o ming g ow h ac o
alpha, a well-cha ac e ized g ow h ac o which plays a key
neu o ophic ole in he cen al and pe iphe al ne ous
sys ems26, and is up egula ed du ing he acu e inju y esponse o
mo o neu ons, p omo ing neu onal su i al27,28.
Th ee lead a ian s o g ip s eng h map in o nea genes
implica ed in monogenic synd omes cha ac e ized by neu ological
and/o psychomo o impai men (Table 1; Supplemen a y Fig. 2).
s10186876 (P
combined
¼9.75 1011)lies15kbups eamo
LRPPRC (leucine- ich pen aco ipep ide-con aining), which has
been implica ed in he F ench-Canadian a ian o Leigh Synd ome
(MIM: 220111), a cy och ome C oxidase deficiency wi h ea u es
including de elopmen al delay, hypo onia and weakness29.
Mu a ions in PEX14 (Pe oxisomal Biogenesis Fac o 14) (in onic
lead a ian s6687430) unde lie ce ain o ms o Zellwege
Spec um Pe oxisomal Biogenesis Diso de (MIM: 614887), a
synd ome cha ac e ized by absence o unc ional pe oxisomes
and sys emic neu ological impai men 30. Finally, s80103986 is
in onic in KANSL1, which has been implica ed in he complex
impai ed-psychomo o pheno ype o Koolen-de V ies synd ome
(MIM: 610443)31. Fu he , he signal a KANSL1 is in a la ge
linkage disequilib ium (LD) block also con aining MAPT
( s754512, P
disco e y
¼3.7 108), which encodes he
mic o ubule-associa ed au p o ein. MAPT has been implica ed in
a sui e o so-called auopa hies cha ac e ized by p og essi e
neu ological defici , and is also a isk locus o Pa kinson’s
disease32. 17q21.31 has a complex haplo ype s uc u e comp ising
an in e sion and h ee s uc u al copy numbe a ian s a ising
om duplica ion e en s, which has p e iously been shown o be o
ele ance o heal h33. A e impu ing he nine common s uc u al
haplo ypes a his locus33–35, haplo ype was significan ly associa ed
wi h g ip s eng h. In pa icula , he in e ed haplo ype was
associa ed wi h lowe s eng h (b¼0.17 kg, P¼3.85106),
independen o age, sex, heigh and BMI (Supplemen a y
Table 3A). This associa ion appea ed o be d i en by he in e ed
a2.g2 s uc u al a ian (b¼0.18kg, P¼1.24105).
In sensi i i y analyses, we e- es ed associa ions o he 16 g ip
s eng h a ian s in UKB a e exclusion o up o 8,676
indi iduals wi h ype 1 diabe es, cance , o o he p e alen
disease wi h po en ial o influence muscle s eng h. No loci
showed significan a enua ion o e ec ela i e o o e all analyses
(Supplemen a y Table 3B and C).
Signals a e en iched o biologically ele an issues. To iden i y
en ichmen o associa ion signals ac oss di e en issues and
iden i y likely e ec o issues, we pe o med cell ype-specific
pa i ioned he i abili y36 analyses on genome-wide associa ion
esul s om he disco e y phase. A e adjus men o mul iple
es ing ac oss nine dis inc issue ypes (Po0.0056), we obse ed
significan en ichmen s o associa ions wi h g ip s eng h in
issue-specific egula o y egions o a numbe o issues,
including bone/connec i e issue (P¼2.03 1010), skele al
muscle (P¼1.88 109) and he CNS (P¼7.37 10 8).
En ichmen s a weake le els o s a is ical significance we e also
obse ed in ca dio ascula and gas oin es inal issue, as well as
he ad enal/panc eas axis, and ‘o he ’ issues (Supplemen a y
Fig. 4).
In eg a ion o gene exp ession da a. Guided by issue-specific
en ichmen s, we sough o iden i y pu a i e e ec o ansc ip s
unde lying hese associa ions by in es iga ing associa ions o
lead SNVs o hei p oxies ( 240.8) wi h ansc ip le els in b ain,
ibial ne e and skele al muscle in GTEx (Supplemen a y Table 4).
The g ip s eng h-inc easing allele a ACTG1 ( s6565586) was
S a a n
<50 yea s
50–60 yea s 37,543
50,326
23,991
Be a (95% CI) P- alue
4.00 × 10–24
1.91 × 10–24
4.47 × 10–44
2.38 × 10–48
3.61 × 10–43
0.19 (0.15, 0.23)
0.15 (0.12, 0.17)
0.17 (0.14, 0.19)
0.20 (0.18, 0.23)
0.13 (0.11, 0.15)
53,090
58,770
Be a (kg pe g ip s eng h-inc easing allele)
–0.10 –0.05 0.00 0.05 0.15 0.25 0.300.200.10
>60 yea s
Male
Female
Figu e 1 | Associa ion o he 16 SNV g ip s eng h sco e wi h g ip s eng h by age and sex s a a. Associa ion o he g ip s eng h-inc easing gene ic
sco e showed no in e ac ion wi h obse ed g ip s eng h by age (p
in e ac ion
¼0.30) bu was s onge in men han in women (P
in e ac ion
¼1.56 105)ina
subse o 111,860 un ela ed UK Biobank pa icipan s om s age one analyses. Associa ions shown a e om linea eg ession.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015 ARTICLE
NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions 3
associa ed wi h lowe exp ession o ACTG1 in skele al muscle
(P
EXP-Lead
¼3.81 1013,P
EXP-Bes eQTL
¼2.64 10 13, 2¼0.85),
coun e odi ec ionssugges edbyac g1 knockou in mouse. A
LRPPRC ( s10186876), he s eng h-inc easing allele was associa ed
wi h highe LRPPRC exp ession le els in ce ebellum
(P
EXP-Lead
¼6.35 107,P
EXP-Bes eQTL
¼9.35 108, 2¼0.93)
and ce ebella hemisphe e (P
EXP-Lead
¼1.29 106,
P
EXP-Bes eQTL
¼3.62 108, 2¼1.00), which appea s di ec ionally
conco dan wi h p e iously cha ac e ized loss o unc ion and
o he wise damaging mu a ions associa ed wi h he disease pheno ype
o F ench-Canadian Leigh Synd ome37. A s1161433 (ERP27 locus),
he g ip s eng h-inc easing allele was associa ed wi h highe le els o
MGP exp ession in ibial ne e (P
EXP-Lead
¼5.90 1010,
P
EXP-Bes eQTL
¼2.19 1012, 2¼0.84). MGP (Ma ix Gla
P o ein) is a well-cha ac e ized inhibi o o ascula issue and
ca ilage calcifica ion, and consequen lyac sasakey egula o o
bone o ma ion38.
We also pe o med in eg a ed ansc ip ome-wide analyses using
ecen ly-desc ibed Me aXcan39 and SMR app oaches40.
Accoun ing o 5973 independen exp ession p obes (Bon e oni-
co ec ed P 8.37 106 o a 0.05), and po en ial coinciden al
o e lap o eQTL signals wi h GWAS loci, SMR analyses using
whole blood ansc ip ome da a41 sugges ed co ela ion be ween
highe g ip s eng h and lowe exp ession le els o ERP27
(P
SMR
¼2.50 109)andKANSL1 (P
SMR
¼3.05 107), bo h
o which a e implica ed genes om ou GWAS analysis.
Me aXcan analysis iden ified 25 p o ein-coding ansc ip s
implica ed in g ip s eng h a Bon e oni-co ec ed significance in
a leas one o wel e biologically ele an issues om he GTEx
esou ce (neu onal, muscle, connec i e, and ogenic issues and
whole blood; Supplemen a y Table 5). T ansc ip s showed
conco dan ly al e ed exp ession ac oss a numbe o hese
candida e issue ypes (Fig. 2). Fo LRPPRC, o example, we
obse ed associa ion o highe exp ession le els ac oss a numbe
o b ain issue ypes, ibial ne e, whole blood and es is, wi h
highe g ip s eng h. Highe MAPT exp ession in mul iple
b ain egions known o be implica ed in mo o coo dina ion
(co ex, ce ebellum and ce ebella hemisphe e) was also
associa ed wi h highe g ip s eng h.
Pa hways unde lying a ia ion in g ip s eng h. Hypo hesis- ee
gene se en ichmen analysis (GSEA) based on gene-se s o
common unc ional anno a ion, o belonging o p e-defined
canonical pa hways (Supplemen a y Table 6A and B), indica ed
fi e- old en ichmen o associa ion in/nea genes implica ed in
‘posi i e egula ion o p o ein ca abolic p ocess’ (gene on ology
(GO): 1903364, alse disco e y a e (FDR) ¼0.026), and
nominal en ichmen o associa ions nea genes implica ed in
‘dual excision epai in global genomic nucleo ide excision epai ’
(Reac ome: R-HSA-5696400, FDR ¼0.047). Gi en he iden ifi-
ca ion o es ablished psychomo o disease loci amongs index
a ian s o g ip s eng h, we addi ionally in e oga ed ou
associa ion esul s o en ichmen o genes known o be
implica ed in monogenic myopa hies and dys ophies (Suppleme-
n a y Table 6B). G ip s eng h associa ions we e nominally
en iched in he myopa hy-linked gene se (P¼0.017), bu no a
loci implica ed in dys ophic condi ions (P¼0.47).
Insigh s in o o e lap wi h p o-a ophic signalling. Myokine
signalling ia ac i in ype II ecep o s (Ac RII) has been
ecognized as a key pa hway by which muscle mass migh be
p ese ed in many clinical con ex s42,43; ye , we saw no e idence
o en ichmen o associa ions a ound genes in a cus om-defined
pa hway o myos a in/ac i in signalling h ough Ac RII ( e . 42;
Supplemen a y Table 6A and B; desc ibed in mo e de ail in he
me hods sec ion). We also pe o med gene-based associa ion
analyses using VEGAS44 o genes encoding ecep o s and ligands
in he Ac RII signalling pa hway, as well as known a ophy
e ec o s (Supplemen a y Table 7). Two genes (ACVR2B and
FBXO32) showed a significan associa ion wi h g ip s eng h
(Po0.0071, accoun ing o se en gene-based es s; Suppleme-
n a y Table 7). ACVR2B, o which we ound he s onges
e idence o gene-based associa ion (P¼0.0002), encodes he
p incipal ansmemb ane ecep o o myos a in: he a ge o
BYM338, a monoclonal an ibody-based inhibi o o Ac RIIB,
which has shown ea ly p omise in e e sing muscle a ophy and
p omo ing hype ophy in phase I ials45,46.FBXO32 encodes
he E3 ubiqui in ligase A ogin-1/MAFbx, which is ecognized
as undamen al e ec o o a ophy42.
T ans o med ib oblas s
Ce eb al co ex
Ce ebellum
Ce ebella hemisphe e
Basal ganglia (NAcc)
Basal ganglia (cauda e)
Hypo halamus
Hippocampus
Basal ganglia (pu amen)
Whole blood
Tes is
Tibial ne e
Skele al muscle
ASB8
RAET1G
ARL17A
LRRC37A
LRRC37A2
WDR73
LRPPRC
SLC35E2B
KIF1B
SPPL2C
ARHGAP27
ACTG1
CHRDL2
PLEKHM1
H2AFJ
ARHGDIB
CRHR1
FMNL1
MAPT
ERP27
GFAP
MGP
RP11−182J1.16
SLIT1
−7
−4.94
−1.96
0
1.96
4.94
7
Figu e 2 | Me aXcan-p edic ed associa ion o p edic ed gene ansc ip
le els wi h g ip s eng h ac oss biologically ele an issues in GTEx.
Da a a e shown o all genes a which al e ed ansc ip ion was significan ly
associa ed wi h g ip s eng h in a leas one biologically ele an issue,
a e accoun ing o mul iple es ing. Da a a e z-sco es o ansc ip le el
associa ion wi h highe handg ip s eng h, clus e ed by issue. Di ec ion o
z-sco e indica es whe he highe o lowe gene exp ession is associa ed
wi h highe g ip s eng h. Absolu e z-sco e41.96 indica es nominal
significance a P 0.05, and Z4.94 indica es significance a e adjus men
o mul iple es ing (P 7.91 107). NAcc, nucleus accumbens.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015
4NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions
Implica ion o loci in eli e a hle ic pe o mance. Whils poo
g ip s eng h is no mally conside ed a ma ke o ail y, we also
in es iga ed he ole o g ip s eng h-associa ed SNVs in he
opposi e ex eme o physiology: eli e a hle e s a us. We examined
he associa ion o g ip s eng h-associa ed SNVs wi h odds
o being an eli e sp in /powe a hle e in a me a-analysis o
ou s udies o sp in and powe a hle es (N
a hle es
¼616,
N
con ols
¼1,610; see Me hods and Supplemen a y No e o
u he de ails o me hods and pa icipan s). Among he 14
a ailable SNVs, we saw no e idence o associa ion wi h eli e
a hle e s a us (Supplemen a y Table 8). P e iously, a nonsense
mu a ion (R577X) in ACTN3, which encodes he ac in-binding
p o ein a-ac inin 3 in skele al muscle, has been associa ed
wi h eli e a hle e s a us47. We ound a nominally significan
associa ion o he s op-gain a ian (T allele) wi h lowe g ip
s eng h (addi i e: b¼0.062 kg, P¼0.018) (Supplemen a y
Table 9) al hough we ound no e idence o any depa u e
om an addi i e gene ic model a his locus (P
domde
¼0.72;
Supplemen a y Fig. 5).
Va ian associa ion wi h muscle his ology. We also examined
whe he he lead SNVs a g ip s eng h-associa ed loci we e
associa ed wi h in e se-no malized muscle fib e ype and capil-
la y densi y in a small sample in which muscle fib e ype his-
ology and genome-wide geno yping we e a ailable (13 o 16
SNVs we e a ailable; Supplemen a y Table 10). Allowing o 13
es s (Po3.8 103), he g ip s eng h- aising allele a TGFA
was associa ed wi h a lowe p opo ion o ype I (slow- wi ch
oxida i e) muscle fib es (b¼0.16, P¼3.3 103), and a
endency owa ds highe p opo ion o ype IIB ( as - wi ch
glycoly ic) muscle fib es (b¼0.16, P¼5.5 103) (Supplemen-
a y Table 10). We acknowledge limi ed powe in his small
sample o iden i y modes e ec sizes. Gi en a mino allele e-
quency o 0.1 and a sample size o 656 indi iduals, we es ima ed
80% powe o de ec an e ec size o B0.35 SDs.
MR o in e media e pheno ypes on muscle s eng h. Gi en he
oles o sex- and g ow h ho mones and ela ed pheno ypes in
muscle g ow h and de elopmen 48–50, we pe o med summa y
s a is ic MR51,52 o es whe he gene ically-de e mined sex
ho mone binding globulin (SHBG), dehyd oepiand os e one
sulpha e (DHEA-S), insulin and insulin-like g ow h ac o -I
(IGF-I) le els we e associa ed wi h g ip s eng h. Using genome-
wide significan ly associa ed SNVs o SHBG ( e . 53), DHEA-S
( e . 54) and IGF-1 le els55, we saw no e idence o a causal
associa ion wi h g ip s eng h (Supplemen a y Table 11). We saw
some indica ion o causali y o insulin esis ance and as ing
insulin le els in g ip s eng h (Supplemen a y Table 11) in
in e se- a iance and median-weigh ed analyses, al hough he
conside able he e ogenei y in in e se- a iance weigh ed esul s
wa an s a cau ious in e p e a ion (Supplemen a y Table 11;
Supplemen a y Fig. 6).
Muscle s eng h as a possible causal exposu e. Using a MR
app oach, we in es iga ed he po en ially causal ole o muscula
s eng h in bo h mo ali y and disease ou comes, u ilizing he 16
eplica ed loci as an ins umen al a iable o model gene ically
de e mined g ip s eng h as a p oxy o wide muscula s eng h.
Mo ali y: We ound no e idence o a causal ela ionship
be ween muscula s eng h and all-cause mo ali y in 21,043
pa icipan s (5,699 dea hs) d awn om he EPIC-No olk coho
(haza d a io (HR) pe kg highe g ip s eng h (95% CI): 0.96
(0.91, 1.03), P¼0.265) (Fig. 3a). Howe e , gi en wide CIs we also
sough o imp o e powe using a ecen ly published app oach
le e aging da a on pa en al li espan56. Using his app oach in
T ai T ai P- alue P- alue
0.953
0.086
0.229
0.265
0.96 (0.91, 1.03) 0.00 (–0.05, 0.05)
0.04 (–0.01, 0.09)
0.02 (–0.01, 0.05)
0.074
0.878
0.07 (–0.01, 0.15)
0.01 (–0.15, 0.18)
–0.1 0.0 0.1 0.2
Be a (kg·m–2)
Odds a io
0.90 0.95 1.00 1.05
0.90 0.95 1.00 1.05
–0.1 0.0 0.1 0.2
Be a (s anda d de ia ion)
Haza d a io
E ec (95% CI) E ec (95% CI)
1.01 (0.98, 1.04)
1.00 (0.96, 1.03)
1.00 (0.98, 1.03)
0.99 (0.94, 1.03) 0.631
0.433
0.020
0.98 (0.93, 1.03)
0.95 (0.90, 0.99)
0.504
0.859
0.739
Fo ea m BMD
Lumba spine BMD
Femo al neck BMD
Mo ali y (EPIC-No olk)
Pa e nal li espan (UKB)
Pa en al li espan (UKB)
Any ac u e
Myoca dial in a c ion
Co ona y hea disease
Lean mass index
Fa mass index
184,305
171.876
120,326
21,043
133,123
138,096Ma e nal li espan (UKB)
n
ba
dc
Figu e 3 | Mendelian andomiza ion es ima es o he associa ion o g ip s eng h wi h mo ali y and mo bidi y ou comes. (a) Mo ali y and pa en al
li espan in UKB and EPIC-No olk; (b) o ea m bone mine al densi y (BMD), lumba spine BMD and emo al neck BMD in GEFOS; (c) co ona y hea
disease and myoca dial in a c ion in CARDIoGRAMplusC4D, and ac u e isk in GEFOS þEPIC-No olk; (d) lean mass index and a mass index in he
Fenland S udy þEPIC-No olk (n¼12,851). E o ba s eflec 95% CI.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015 ARTICLE
NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions 5
UKB (102,072 pa e nal dea hs, 83,315 ma e nal dea hs), we
again ound no e idence o causali y, wi h g ea e p ecision
(HR (95% CI): 1.00 (0.98, 1.03), P¼0.739) (Fig. 3a).
Co ona y hea disease. We nex in es iga ed a causal ole o
g ip s eng h in ca dio ascula disease using genome-wide
associa ion esul s o co ona y hea disease (CHD; 60,801 cases,
123,504 con ols) and myoca dial in a c ion (MI; 43,677 cases,
128,199 con ols) om he CARDIoGRAMplusC4D Conso -
ium57 (Fig. 3c). We ound no e idence o a causal ela ionship
be ween g ip s eng h and CHD (odds a io (OR) pe gene ically
p edic ed kg highe g ip s eng h (95% CI): 0.99 (0.94, 1.03),
P¼0.631) o MI (OR (95% CI): 0.98 (0.93, –1.03), P¼0.433)
(Supplemen a y Table 12). This esul was u he suppo ed by
c oss- ai LD Sco e eg ession esul s showing no significan
gene ic co ela ion be ween g ip s eng h and CHD (
g
¼0.045,
P¼0.362; Supplemen a y Table 13).
F ac u e isk and bone mine al densi y: We pe o med
MR analyses o ac u e isk using a me a-analysis o 1)
summa y s a is ic MR esul s o ac u e isk om he Gene ic
Fac o s o Os eopo osis (GEFOS) conso ium (ncases ¼20,439;
ncon ols ¼78,843) (Supplemen a y No e; Supplemen a y
Table 12), and 2) logis ic eg ession esul s om associa ion o
he weigh ed g ip s eng h gene ic sco e wi h ac u e isk in he
EPIC-No olk s udy (1,002 cases, 20,042 con ols). Me a-analysis
esul s sugges ed a po en ial causal associa ion o gene ically-
p edic ed highe g ip s eng h wi h lowe isk o ac u e (OR pe
gene ically p edic ed kg highe g ip s eng h (95% CI): 0.95 (0.90–
0.99), P¼0.02; Fig. 3b). Summa y s a is ic MR o gene ically
de e mined g ip s eng h on publicly a ailable bone mine al
densi y (BMD) GWAS esul s58 did no show significan
associa ions be ween g ip s eng h and BMD (Fig. 3c;
Supplemen a y Table 12). Howe e , we did find genome-wide
gene ic co ela ions o bone mine al densi y wi h g ip s eng h
( emo al neck BMD:
g
¼0.123, P¼9.5 103; lumba spine
BMD:
g
¼0.156, P¼6104) (Supplemen a y Table 13),
suppo i e o a ole o gene ically p edic ed g ip s eng h in
ac u e isk.
Simila o he BMD esul s, MR analyses o gene ically-
de e mined g ip s eng h on me a-analysed GWAS esul s
comp ising 12,851 pa icipan s om he Fenland and EPIC-
No olk coho s did no show significan associa ions be ween g ip
s eng h and lean mass index (LMI) o a mass index (FMI) (LMI:
b¼0.072 kg m2,P¼0.074; FMI: b¼0.013 kg m 2,P¼0.878)
(Fig. 3d; Supplemen a y Table 12). A significan gene ic co ela ion
was obse ed be ween g ip s eng h and LMI (
g
¼0.258, P¼2.8
105), bu no FMI (Supplemen a y Table 13).
Discussion
We ha e iden ified 16 loci associa ed wi h maximal hand g ip
s eng h a genome-wide significance. A numbe o he lead
a ian s we e loca ed wi hin o close o genes implica ed in
s uc u e and unc ion o skele al muscle fib es, neu onal
main enance and signal ansduc ion in he cen al and
pe iphe al ne ous sys ems. Pa i ioned he i abili y analyses
indica ed significan issue-specific en ichmen o skele al muscle,
CNS, connec i e issue and bone in he genome-wide g ip
s eng h esul s. We obse ed e idence o sha ed gene ic ae iology
be ween lean mass and g ip s eng h, while pa hway analyses
indica ed a ole o genes in ol ed in egula ion o p o ein
ca abolism in he ae iology o g ip s eng h.
Due o he well-es ablished obse a ional associa ions o g ip
s eng h wi h mo ali y and inciden CHD i has been hypo hesized
ha imp o emen o muscle s eng h migh inc ease longe i y and
educe isk o ad e se ca dio ascula e en s7. Ou MR analyses do
no find e idence suppo i e o a causal ole o muscula s eng h in
mo ali y isk, no in isk o ca dio ascula e en s (CHD and MI),
lea ing open he possibili y ha hese obse a ional associa ions
may be a ibu able o con ounding and/o e e se causali y.
Rega dless, his does no nega e he impo ance o main aining
s eng h and muscle mass du ing ageing as a s a egy o main ain
physical unc ion59, and we acknowledge he po en ial limi a ions o
ou MR. Fo example, he limi ed a iance in in e media e ai s
explained by gene ic a ian s lea es unce ain y o e he p esence o
a small causal e ec . Thus, expanded gene ic disco e y e o s and
g ea e a ailabili y o la ge-scale s udies o disease ou comes will
imp o e he p ecision o MR analyses in u u e. We saw e idence
o sha ed gene ic ae iology o bone mine al densi y and lean mass
wi h g ip s eng h, and MR esul s sugges ed a causal ole o highe
muscula s eng h in lowe isk o ac u e. Collec i ely, hese
esul s sugges ha he de e minan s o muscula s eng h a e
sha ed wi h he de e minan s o ac u e isk and a e consis en
wi h findings om in e en ion s udies o inc ease muscle s eng h,
which ha e been shown o imp o e unc ional capaci y and educe
he a eo alls
59, as well as a enua ing he a e o unc ional decline
and inc eased ail y which o en ollows majo ac u e among he
elde ly60. Despi e he es ablished decline in g ip s eng h wi h
inc easing age, we did no obse e he e ogenei y in he e ec o g ip
s eng h-associa ed a ian s wi h g ip s eng h by age. Howe e , we
did obse e ha he cumula i e e ec o he gene ic sco e was
g ea e in men han women.
We saw e idence o en ichmen o associa ions wi h g ip
s eng h a ound genes implica ed in myopa hies. We also no ed
h ee loci wi h genome-wide significan associa ions con aining
genes (KANSL1, PEX14 and LRPPRC) implica ed in a e, se e e
clinical synd omes cha ac e ized by pheno ypes o p og essi e
psychomo o impai men , muscle hypo onia and neu opa-
hy29,31. Wi hin he UKB disco e y coho , he associa ion o
all h ee loci wi h g ip s eng h pe sis ed a genome-wide
significance e en a e sensi i i y analyses es ic ed o
pa icipan s wi hou any o m o sel - epo ed condi ion which
migh a ec muscle mass o unc ion. Whils hese clinical
condi ions ep esen he ex eme pheno ype o highly dele e ious
a e mu a ions in hese genes, we demons a e ha p oximal
common a ian s a e likely o unde pin mo e sub le popula ion-
le el a ia ion in s eng h in heal hy popula ions.
Finally, we ound ha common a ia ion a ACVR2B, he
p incipal ecep o o myos a in and ac i in in skele al muscle, is
associa ed wi h popula ion-le el a ia ion in g ip s eng h. Ongoing
clinical ials and de elopmen o pha maceu ical agen s a ge ing
his pa hway ha e demons a ed hei po en ial o e e se a ophy
and imp o e physical unc ioning46, and ou findings p o ide some
le el o gene ic suppo o a ole in muscula s eng h.
In conclusion, we iden ified 16 loci obus ly implica ed in g ip
s eng h and p o ide insigh in o he unde lying biology o his
impo an , widely s udied, ye poo ly cha ac e ized ai . MR
analyses sugges no causal ole o muscula s eng h in mo ali y,
bu do p o ide e idence o a causal ole in ac u e isk,
highligh ing he impo ance o in e en ions o imp o e muscle
s eng h as a means o educe ac u e isk and esul an
mo bidi ies. Fu he gene ic and unc ional wo k o cha ac e ize
hese loci will elucida e new pa hways in ol ed in he egula ion
o muscle s eng h and in o m he de elopmen o d ugs o ackle
muscle was ing and weakness.
Me hods
S udy coho s.S age one (disco e y) analyses comp ised pa icipan s d awn om
he UK Biobank (UKB) s udy17, a la ge popula ion-based coho o middle and
olde -aged (40–69 yea s) B i ish esiden s ec ui ed om UK Na ional Heal h
Se ice (NHS) p ima y ca e egis e s be ween 2006 and 2010. In o al, 503,325
pa icipan s we e en olled, and a ended an ini ial assessmen isi a one o
22 s udy cen es loca ed h oughou England, Sco land and Wales, du ing which a
comp ehensi e ca alogue o an h opome ic, li es yle and beha iou al exposu es
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015
6NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions
we e assessed, and biological samples we e a ained. All pa icipan s p o ided
in o med consen . UKB gained e hical app o al om he Na ional Resea ch E hics
Commi ee (No h Wes ) and was conduc ed in ull compliance wi h p inciples o
he Wo ld Medical Associa ion Decla a ion o Helsinki.
Independen lead a ian s om s age one we e ollowed-up (s age wo analyses)
in an independen sample o up o 53,145 whi e Eu opean indi iduals d awn om
he Coho s o Hea and Aging Resea ch in Genomic Epidemiology (CHARGE)
conso ium16 and an addi ional se en collabo a ing s udies, which had assessed
maximal isome ic hand g ip s eng h by dynamome y. De ails o all s age wo
coho s (including he cons i uen coho s o CHARGE) a e p o ided in he
Supplemen a y No e. Fu he desc ip i e de ails o he se en addi ional coho s a e
p o ided in Supplemen a y Table 1. Desc ip i e de ails o CHARGE Coho s ha e
been published in de ail elsewhe e16.
Assessmen o g ip s eng h and co a ia es.Hand g ip s eng h a baseline in
UKB was measu ed isome ically using a calib a ed Jama J00105 hyd aulic
hand dynamome e (La aye e Ins umen Company, IN, USA) adjus ed o he
indi idual’s hand size. Wi h he pa icipan sea ed, one measu emen was aken
pe hand, and maximal g ip s eng h aken as he highe o he wo eadings.
Body mass was assessed using a BC418MA Body Composi ion Analyse
(Tani a Eu ope BV, Ams e dam, The Ne he lands), wi h he pa icipan d essed
in ligh clo hing. S anding heigh was measu ed on a igid s adiome e
(Seca, Bi mingham, UK). Pheno yping de ails o each o he s age wo coho s
a e de ailed in Supplemen a y Table 1.
Geno yping and impu a ion.Ou s age one analyses use da a om UKB’s
impu ed in e im geno yping elease (May 2015), es ic ed o biallelic SNVs wi h
MAF Z0.1%. Geno yping and impu a ion we e conduc ed by UKB using a
cen alized pipeline, o which de ailed p o ocols a e a ailable (see URLs). B iefly,
UKB ex ac ed DNA om EDTA bu y coa , be o e shipping o A yme ix
(San a Cla a, CA, USA) o cen alized geno yping. Samples we e geno yped
a 4800,000 loci on wo cus om-designed a ays wi h 95% common con en ,
designed o op imize quali y and quan i y o genome-wide impu a ion: he UK
Biobank Axiom a ay, and he UK BiLEVE Axiom A ay. A e es ic ion o
biallelic SNVs wi h MAFZ1% and addi ional sample geno yping QC, a subse o
641,081 au osomal SNVs om 152,256 samples we e a ailable o impu a ion.
SNVs we e p e-phased using SHAPEIT3 so wa e and impu ed (using a modified
e sion o IMPUTE2 so wa e) o a me ged e e ence panel con aining haplo ypes
om he UK10K Conso ium combined wi h he 1000 Genomes P ojec e e ence
(see URLs). This app oach has p e iously been shown o p o ide a high-quali y
impu a ion e e ence in popula ions o mixed ances y61. Geno yping and
impu a ion de ails o s age wo coho s a e de ailed in Supplemen a y Table 1.
17q21.31 Haplo ype impu a ion and analyses.Nine s uc u al haplo ypes
p e iously epo ed a 17q21.31 we e impu ed acco ding o p e ious wo k33–35.
Impu a ion was based on a haplo ype e e ence panel34, which uniquely coded each
s uc u al haplo ype using a combina ion o wel e su oga e, i ual bina y
ma ke s. In addi ion, he file con ained 6,302 flanking a ian haplo ypes. IMPUTE
2.3.2 was used o impu e he geno ypes o he su oga e ma ke s agains he
e e ence panel. The panel con ained 284 geno yped a ian s wi hin he e e ence
egion, p e-phased wi h SHAPEIT 2.837. Va ian s wi hin he copy-numbe
a iable egion, o wi h MAFo0.01/Ha dy-Weinbe g Equilib ium Po110 6
we e excluded. Su oga e ma ke s we e subsequen ly decoded in o he
co esponding nine s uc u al haplo ypes o analysis. Associa ion o haplo ypes
(in e ed e sus non-in e ed, con inuous s uc u al a ian [a/b/g] copy numbe ,
and 9 common haplo ypes as a ca ego ical exposu e) wi h g ip s eng h was
modelled using linea eg ession adjus ed o age, sex, heigh (m), BMI (kg m 2)
and UKB geno ype chip in up o 111,860 un ela ed gene ic whi e Eu opeans
defined cen ally by UKB (see URLs).
He i abili y es ima ion.In UKB, a iance componen analyses we e pe o med in
he subse o indi iduals o ‘whi e B i ish’ gene ic ances y using Res ic ed Es i-
ma e Maximum Likelihood (REML) models in BOLT-LMM so wa e ( 2.2)62.
Gene ic a iance was calcula ed on all quali y con olled geno yped au osomal
SNVs, adjus ing o geno yping a ay and he op fi e gene ically-de e mined
p incipal componen s.
Genome-wide associa ion analyses o g ip s eng h.142,035 UKB pa icipan s
had impu ed gene ic da a, g ip s eng h and ull co a ia e a ailabili y o
genome-wide associa ion analyses; all we e o sel -iden ified whi e ances y, wi h
he majo i y (94.6%) epo ing as whi e B i ish. Disco e y analyses o maximal
g ip s eng h (n¼142,035) we e un using a Bayesian linea mixed model (LMM)
adjus ed o age (yea s), sex, heigh (m) and BMI (kg m2), implemen ed in
BOLT-LMM so wa e ( 2.2)62. P ima y analyses assumed addi i e (pe -allele)
e ec . Analyses we e es ic ed o biallelic a ian s wi h MAFZ0.1% which had
been di ec ly yped, o impu ed wi h impu a ion quali y (IMPUTE2 in o)Z0.4.
LMMs o e a obus solu ion o handle unknown con ounding (pa icula ly
ha a ising om sub-e hnic popula ion s a ifica ion and c yp ic ela edness)
in genome-wide associa ion s udies, and con e inc eased powe in la ge
popula ion-based coho s62.
Independen loci om genome-wide disco e y we e defined as he 500 kb
egion flanking each lead a ian eaching genome wide significance
(P 5108). Independen lead a ian s (n¼21) we e ollowed-up in up o
53,145 indi iduals. In cases whe e an index a ian was no yped o impu ed a
su ficien quali y, app op ia e p oxies we e defined as he a ian wi h he
nex -lowes P- alue wi hin 500 kb o he index (Supplemen a y Table 15). Each
s age wo coho accoun ed o popula ion s uc u e acco ding o i s usual p ac ice.
Full de ails o he analy ical app oach and model specifica ion o each eplica ion
coho a e summa ized in Supplemen a y Table 1. S age one and s age wo esul s
o each o he 21 a ian s we e combined by in e se a iance-weigh ed fixed-e ec
me a-analysis using METAL. Six een loci eached P 5108in combined
me a-analysis and we e conside ed o be associa ed wi h g ip s eng h.
To examine local linkage disequilib ium s uc u e o eplica ed loci, egional
plo s o each o he 16 eplica ed loci we e gene a ed in LocusZoom using LD
e e ence alues om he CEU panel o 1000G Phase I. To in es iga e possible
independen signals a each o he 16 eplica ed loci, app oxima e condi ional
analyses we e unde aken using Genome-Wide Complex T ai Analysis so wa e
(GCTA, Ve sion 1.25.2).
LD sco e eg ession.Using genome-wide summa y s a is ics om ou UKB phase
one analyses, he ecen ly-desc ibed LD Sco e Reg ession me hod desc ibed by
Bulik-Sulli an and colleagues63 (implemen ed in LDSC so wa e, 1.0.0) was used
o (i) es ima e gene ic co ela ion be ween g ip s eng h and o he pheno ypes, and
(ii) de i e issue-specific pa i ioned he i abili y o g ip s eng h, based on
p e-calcula ed Eu opean LD Sco es. To a oid con ounding by impu a ion quali y,
all analyses we e es ic ed o a ian s a ailable in HapMap Phase III.
Gene ic co ela ions. Using c oss- ai LD Sco e eg ession, genome-wide gene ic
co ela ions o g ip s eng h we e calcula ed wi h CHD isk, lean mass index (LMI),
a mass index (FMI) and bone mine al densi y (BMD) measu ed a he o ea m,
emo al neck o lumba spine. Fo CHD and BMD we used publicly a ailable GWAS
summa y s a is ics o m he CARDIoGRAMplusC4D57 and GEFOS58 conso ia,
espec i ely. To ob ain genome-wide summa y s a is ics o LMI and FMI, we
conduc ed GWAS in up o 12851 indi iduals d awn om he Fenland S udy and
EPIC-No olk. De ails o hese coho s a e p o ided in he Supplemen a y No e, and
Supplemen a y Table 14. LMI and FMI (kg m2)we edefinedasdualx- ay
abso p iome y (DXA)-de i ed lean mass o a mass, espec i ely, di ided by he
squa e o DXA-de i ed heigh (GE Luna P odigy p ocessed using Luna EnCORE
14.1, GE Heal hca e). GWAS we e conduc ed sepa a ely in each coho unning a
linea mixed model using BOLT-LMM62. FMI analyses we e adjus ed o age and sex.
Because o he sex specific dis ibu ion o he pheno ype, LMI analyses we e un sex
s a ified and adjus ed o age. Resul s om bo h coho s we e combined by fixed-
e ec in e se a iance-weigh ed me a-analysis using METAL.
Tissue-specific pa i ioned he i abili y: Pa i ioned he i abili y by LD Sco e
Reg ession can iden i y whe he ce ain cell ypes a e en iched o unc ional gene
ca ego ies which disp opo iona ely con ibu e o he he i abili y o a pheno ype36.
In his way, o e - ep esen ed e ec o issues impo an in he ae iology o he
pheno ype can be iden ified. Pa i ioned he i abili y was un indi idually o each
o he eigh cu a ed issue classes dis ibu ed wi h LDSC, adjus ing in each case o
he i abili y explained by each unc ional ca ego y o a ian s ac oss he genome
( ha is, in a non- issue-specific manne ). A Bon e oni-co ec ed log
10
P- alue
accoun ing o eigh es s was aken as indica i e o s a is ical significance.
Exp ession analyses.To explo e he po en ial unc ional significance o g ip
s eng h a ian s in gene exp ession, and o p io i ize unc ional genes alling
wi hin iden ified loci, we unde ook a numbe o exp ession-based analyses.
Ini ially, a look-up o all 16 eplica ed g ip s eng h a ian s o hei bes p oxy
( 240.8) was conduc ed in skele al muscle, ans o med fib oblas s, ne ous
sys em and b ain egions in he GTEx esou ce o iden i y eQTL associa ions
(see URLs). Va ian s passing GTEx c i e ia o issue-specific eQTL associa ion,
and in high LD ( 2Z0.8) wi h he bes eQTL o he ansc ip in ques ion in he
issue o in e es we e conside ed significan eQTLs.
To supplemen his a ian -cen ic app oach, we addi ionally ook ad an age o
wo new me hods o in eg a ing genome wide GWAS summa y s a is ics wi h
exp ession associa ions om independen s udies: SMR40 and Me aXcan39.By
u ilizing es ablished eQTL da a se s as e e ence, hese app oaches a e able o
e ec i ely model expec ed a ia ion in he ansc ip ome o he GWAS sample
based on a ia ion in au osomal SNVs ac oss he genome, and hen es o
independen associa ions be ween impu ed ansc ip le els and he pheno ype o
in e es . To es o associa ions o ansc ip abundance wi h g ip s eng h in
whole blood, we implemen ed SMR so wa e using published whole blood eQTL
da a om Wes a and colleagues41 as he e e ence panel. Me aXcan—an ex ension
o he P ediXcan app oach modified o use summa y-le el associa ion s a is ics as
inpu —analyses we e used o explo e u he issue-specific associa ions be ween
modelled gene exp ession and g ip s eng h. Tissue-specific exp ession p edic ion
models gene a ed om GTEx we e downloaded om he P edic DB esou ce
(see URLs) as ansc ip ome e e ence. We conse a i ely conside ed
p edic ed exp ession o a gene o be associa ed wi h g ip s eng h a a Me aXcan
P- alue 2.52 107, aking each gene associa ion in each issue as an
independen es o he pu poses o Bon e oni co ec ion.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015 ARTICLE
NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions 7
Gene se en ichmen analyses.Genome-wide disco e y esul s om he UKB
coho we e es ed o en ichmen o p e-specified gene se s based on common
unc ional anno a ion o known biological pa hways in MAGENTA ( 2.4)64.
En ichmen was assessed in a hypo hesis- ee manne ac oss gene se s d awn om
six public da abases o gene on ology, unc ional anno a ion and canonical/cu a ed
pa hways: GO Te ms, he P o ein Analysis h ough E olu iona y Rela ionships
(PANTHER) da abase, Ingenui y, he Kyo o Encyclopaedia o Genes and Genomes
(KEGG), Bioca a and Reac ome pa hways (downloaded ia he Molecula
Signa u es Da abase, MSigDB [see URLs], July 2011). Analyses we e es ic ed o
3,216 gene se s wi h an e ec i e size o Z10 genes a e fil e ing o p oximal genes
and genes no con aining a ian s o analysis. Cus om se s we e also defined om
li e a u e e iew, inco po a ing genes wi h known unc ion in pa hways o
p ocesses ele an o muscle de elopmen and main enance. Genes in ol ed in
signal ansduc ion o myos a in/ac i in signalling ia Ac i in A ype II ecep o s
(ACVR2A and ACVR2B) we e defined om Han e al.42. Monogenic genes
implica ed in muscula dys ophies and myopa hies we e based on Kaplan
& Ham oun (2014)65 wi h addi ional manual cu a ion o include collagen
IV-opa hies, congeni al myopa hies and glycogen s o age diseases which may
p esen wi h a simila pa e n o limb-gi dle muscle weakness.
Gene-based associa ion es s.Genes in ol ed in myos a in/ac i in signalling ia
ac i in ype II ecep o s (FST, MSTN, ACVR2A, ACVR2B) and known a ophy
e ec o s TRIM63 and FBXO32 we e iden ified om li e a u e42,43 and defined as
candida e genes o g ip s eng h based on hei biological p io o an in ol emen
in skele al muscle ophism. Gene-based associa ion es s we e pe o med o each
candida e using he Ve sa ile Gene-based Associa ion S udy (VEGAS) algo i hm44,
calcula ing LD om HapMap Eu opeans. VEGAS was applied o he g ip s eng h
disco e y-phase associa ion esul s ac oss he whole genome, es ic ing o di ec ly
yped and well-impu ed a ian s (IMPUTE in o40.8).
Muscle his ology lookup.To iden i y whe he g ip s eng h a ian s we e
associa ed wi h elemen s o muscle his ology, we looked-up each o he 16
eplica ed loci om combined analyses in a p e-exis ing GWAS o muscle his ology
pa ame e s in a sample o 656 men om h ee independen coho s o Swedish
ances y (see Supplemen a y No e o coho de ails). Specifically, we in es iga ed
he linea (addi i e) associa ion o each o he 16 lead SNVs wi h pe cen age o
(i) ype I fib es, (ii) ype IIA fib es and (iii) ype IIB fib es om muscle biopsy, as
well as capilla y densi y (calcula ed as he numbe o capilla ies di ided by he
o al numbe o fib es). Pheno ypes we e in e se-no malized p io o analysis. To
be e quan i y powe in his sample, o mal powe calcula ions we e pe o med
using Quan o (see URLs).
Mendelian andomiza ion analyses.We pe o med summa y s a is ic Mendelian
andomiza ion (MR)51,52 o es whe he gene ically de e mined sex and g ow h
ho mone- ela ed pheno ypes we e causally associa ed wi h g ip s eng h. As
p ima y analyses we pe o med in e se a iance weigh ed summa y s a is ics
MR52. In addi ion, as sensi i i y analyses o obus causal in e ence we es ed o
he e ogenei y using Coch an’s Q es , an MR-Egge 66 o assess o pleio opic
e ec , and addi ionally used a weigh ed median es ima o and penalized weigh ed
median es ima o 67. We used publicly a ailable genome-wide associa ion esul s o
sex ho mone binding globulin (SHBG)53, dehyd oepiand os e one sulpha e
(DHEA-S)54, as ing insulin68, insulin sec e ion69 and insulin-like g ow h ac o -I
(IGF-I)55. The a ian s included in he MRs a e lis ed in Supplemen a y Table 16.
G ip s eng h summa y s a is ics we e ob ained om ou s age one GWAS in UKB.
To in e causali y in he associa ion o g ip s eng h wi h CHD, myoca dial
in a c ion (MI), ac u e isk, BMD ( o ea m, lumba spine, emo al neck), LMI and
FMI, we an summa y s a is ic MR as desc ibed abo e using he 16 iden ified loci as
ins umen al a iable o gene ically-de e mined g ip s eng h. Fo CHD, BMD, LMI
and FMI we used he same GWAS summa y s a is ics as we used o es gene ic
co ela ions (see abo e). In addi ion, we used publicly a ailable MI summa y s a is ics
om he CARDIoGRAMplusC4D conso ium57, ac u e isk summa y s a is ics
om an ongoing analysis by he GEFOS conso ium (Supplemen a y No e), and
indi idual ac u e isk da a in EPIC-No olk. Whe e g ip s eng h lead SNVs we e
no a ailable in he ou come pheno ype summa y s a is ics, p oxies we e defined as
he a ian wi h he nex -lowes P- alue o associa ion wi h g ip s eng h wi hin
500 kb o he index in s age one (UKB). All a ian s included in he analyses a e
de ailed in Supplemen a y Table 18. Because EPIC-No olk was included in he LMI
and FMI GWAS me a-analyses, and in he indi idual le el da a ac u e isk analyses,
we used he g ip s eng h e ec sizes ob ained a e exclusion o he EPIC-No olk
s udy in he g ip s eng h-LMI, g ip s eng h-FMI and indi idual le el g ip s eng h-
ac u e isk MR analyses (Supplemen a y Table 17). On he indi idual le el ac u e
isk da a we an a logis ic eg ession model adjus ed o age and sex. Summa y
s a is ics and indi idual le el ac u e isk MR esul s om GEFOS and EPIC-No olk
we e me a-analysed using fixed e ec s me a-analysis.
To es he causal ela ionship wi h all-cause mo ali y, we calcula ed a gene ic
g ip s eng h isk sco e pe indi idual in he EPIC-No olk s udy (n o al ¼21,043,
ncases ¼5,699 cases) based on he numbe o g ip s eng h-inc easing alleles
weigh ed by he e ec size om he combined phase one and ollow-up analyses.
We used e ec sizes ob ained by fixed-e ec in e se a iance-weigh ed
me a-analysis o he phase one and wo esul s, excluding EPIC-No olk, o
gene a e weigh s ha we e independen o EPIC-No olk (Supplemen a y
Table 17). The gene ic isk sco e o mo ali y associa ion was es ed unde a Cox
p opo ional haza ds model adjus ed o age and sex. P opo ional haza ds we e
confi med using s anda d echnique.
We also sough o imp o e powe by using pa en al li espans in UKB (pa e nal:
n
TOTAL
¼133,123, n
DEATHS
¼102,072; ma e nal: n
TOTAL
¼138,096,
n
DEATHS
¼83,315), in line wi h p e ious wo k56. Pa en al li espans and ali e/dead
s a us we e eg essed using Cox models on o sp ing geno ype, in e ec impu ing
pa en geno ype om o sp ing. The e ec s obse ed hus eflec he e ec o
o sp ing geno ype on pa en al pheno ype, and he expec ed allelic dosages in he
pa en al gene a ion a e hal he measu ed dosages in o sp ing. E ec es ima es pe
pa en al allele a e co espondingly wice ha obse ed pe o sp ing allele: esul s
shown a e he e ec o one allele in pa en s on pa en s’ li espan.
Associa ion o eplica ed loci wi h eli e a hle ic s a us.Using da a om ou
mul i-e hnic coho s o eli e a hle es, including eli e Japanese a hle es and con ols
(N
a hle es
¼54, N
con ols
¼406); eli e A ican-Ame ican ((N
a hle es
¼79,
N
con ols
¼391) and Jamaican sp in /powe a hle es (N
a hle es
¼88, N
con ols
¼87),
and Eu opean a hle es (N
a hle es
¼395, N
con ols
¼726) (Supplemen a y No e), we
assessed he associa ion o he 16 eplica ed g ip s eng h index a ian s wi h odds
o a aining eli e a hle e s a us, ela i e o age, sex and e hnically-ma ched con ols,
using condi ional logis ic eg ession (addi i e model). Analyses we e pe o med
sepa a ely in each coho , and me a-analysed using METAL.
Tes s o model fi .To es o depa u e om addi i i y, we used a es o
dominance de ia ion, including wo e ms o bes guess geno ypes: a e m
encoding he majo homozygo es, he e ozygo es and mino allele homozygo es as
0,1,2 and ano he coding hem as 0,1,0, which es s whe he he he e ozygo es ha e
mean ai alues hal way be ween he homozygo e g oups and can de ec a
depa u e om addi i i y.
Checks o allele selec ion by age.Gi en ha obse a ional g ip s eng h is
s ongly p edic i e o mo ali y8, we an wo complemen a y analyses in UKB o
ensu e ha s eng h-inc easing alleles om combined s age one þ wo analyses
we e no unde selec ion by age. Modelling each SNV as s eng h-inc easing allele
dosage, linea eg ession was used o assess he associa ion o age wi h allele dosage
(age as dependen a iable). We hen pe o med he in e se o his eg ession o
gauge whe he allele dosage was p edic ed by age (age as he independen a iable).
This app oach has ecen ly been applied o es o selec ion o a ian s by age in
he Gene ic Epidemiology Resea ch on Aging (GERA) coho 18. Analyses we e
es ic ed o 112,337 un ela ed whi e Eu opeans defined cen ally by UKB, and
adjus ed o sex and geno yping chip.
URLs.UK Biobank Geno yping and QC Documen a ion h p://biobank.c su.ox.-
ac.uk/c ys al/docs/geno yping_qc.pd ; UK Biobank Phasing and Impu a ion P o-
ocol; h p://biobank.c su.ox.ac.uk/c ys al/docs/impu e_ukb_ 1.pd ; MSigDB;
h p://so wa e.b oadins i u e.o g/gsea/msigdb; SMR; h p://cnsgenomics.com/
so wa e/sm ; P edic DB Da abase; h p://p edic db.hakyimlab.o g; Quan o; h p://
bios a s.usc.edu/Quan o.h ml
Da a a ailabili y.S age one da a a e om UK Biobank, and can be ob ained upon
applica ion (ukbiobank.ac.uk). Access o unde lying eplica ion and ollow-up da a
including his ology and eli e a hle ic pe o mance coho s may be limi ed by
pa icipan consen and da a sha ing ag eemen s; eques s should be di ec ed in he
fi s ins ance ia he co esponding au ho s. P e-defined gene se s (MSigDB),
exp ession da a (GTEx) and ansc ip ome models used by Me aXcan (P edic DB)
and SMR me hods a e a ailable om he lis ed URLs.
Re e ences
1. Bohannon, R. W. Hand-g ip dynamome y p edic s u u e ou comes in aging
adul s. J. Ge ia . Phys. The . 31, 3–10 (2008).
2. Taekema, D. G., Gussekloo, J., Maie , A. B., Wes endo p, R. G. J. & de C aen, A.
J. M. Handg ip s eng h as a p edic o o unc ional, psychological and social
heal h. A p ospec i e popula ion-based s udy among he oldes old. Age Ageing
39, 331–337 (2010).
3. Ran anen, T. e al. Muscle s eng h as a p edic o o onse o ADL dependence
in people aged 75 yea s. Aging Clin. Exp. Res. 14, 10–15 (2002).
4. Cheung, C.-L. e al. Low handg ip s eng h is a p edic o o os eopo o ic
ac u es: c oss-sec ional and p ospec i e e idence om he Hong Kong
Os eopo osis S udy. Age 34, 1239–1248 (2012).
5. Ka
¨ kka
¨inen, M. e al. Associa ion be ween unc ional capaci y es s and
ac u es: an eigh -yea p ospec i e popula ion-based coho s udy. Os eopo os.
In . 19, 1203–1210 (2008).
6. Sa ino, E. e al. Handg ip s eng h p edic s pe sis en walking eco e y a e
hip ac u e su ge y. Am. J. Med. 126, 1068–75.e1 (2013).
7. Leong, D. P. e al. P ognos ic alue o g ip s eng h: findings om he
P ospec i e U ban Ru al Epidemiology (PURE) s udy. Lance 386, 266–273
(2015).
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015
8NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions
8. Coope , R., Kuh, D. & Ha dy, R.Mo ali y Re iew G oup & FALCon
and HALCyon S udy Teams. Objec i ely measu ed physical capabili y
le els and mo ali y: sys ema ic e iew and me a-analysis. BMJ 341, c4467
(2010).
9. O ega, F. B., Sil en oinen, K., Tynelius, P. & Rasmussen, F. Muscula s eng h
in male adolescen s and p ema u e dea h: coho s udy o one million
pa icipan s. BMJ 345, e7279 (2012).
10. Reed, T., Fabsi z, R. R. R., Selby, J. V. V. & Ca melli, D. Gene ic influences and
g ip s eng h no ms in he NHLBI win s udy males aged 59-69. Ann. Hum.
Biol. 18, 425–432 (1991).
11. A den, N. K. & Spec o , T. D. Gene ic influences on muscle s eng h, lean
body mass, and bone mine al densi y: a win s udy. J. Bone Mine . Res. 12,
2076–2081 (1997).
12. Ma eini, A. M. e al. He i abili y es ima es o endopheno ypes o long and
heal h li e: he Long Li e Family S udy. J. Ge on ol. A. Biol. Sci. Med. Sci. 65,
1375–1379 (2010).
13. Saye , A. A. e al. Polymo phism o he IGF2 gene, bi h weigh and g ip
s eng h in adul men. Age Ageing 31, 468–470 (2002).
14. Da o, S. e al. UCP3 polymo phisms, hand g ip pe o mance and su i al a old
age: associa ion analysis in wo Danish middle aged and elde ly coho s. Mech.
Ageing De . 133, 530–537 (2012).
15. Chan, J. P. L. e al. Gene ics o hand g ip s eng h in mid o la e li e. Age 37,
9745 (2015).
16. Ma eini, A. M. e al. GWAS analysis o handg ip and lowe body
s eng h in olde adul s in he CHARGE conso ium. Aging Cell 15, 792–800
ð2016Þ:
17. Sudlow, C. e al. UK Biobank: an open access esou ce o iden i ying he causes
o a wide ange o complex diseases o middle and old age. PLoS Med. 12,
e1001779 (2015).
18. Mos a a i, H., Be isa, T., P zewo ski, M. & Pick ell, J. K. Iden i ying gene ic
a ian s ha a ec iabili y in la ge coho s. P ep in a bioRxi h ps://doi.o g/
10.1101/085969 (2016).
19. E as i, J. M. Cos ame es: he Achilles’ heel o He culean muscle. J. Biol. Chem.
278, 13591–13594 (2003).
20. Rybako a, I. N., Pa el, J. R. & E as i, J. M. The dys ophin complex o ms a
mechanically s ong link be ween he sa colemma and cos ame ic ac in. J. Cell
Biol. 150, 1209–1214 (2000).
21. Dalkilic, I. & Kunkel, L. M. Muscula dys ophies: genes o pa hogenesis. Cu .
Opin. Gene . De . 13, 231–238 (2003).
22. Sonnemann, K. J. e al. Cy oplasmic gamma-ac in is no equi ed o skele al
muscle de elopmen bu i s absence leads o a p og essi e myopa hy. De . Cell
11, 387–397 (2006).
23. Bu , A. R. e al. Na þdys egula ion coupled wi h Ca2 þen y h ough
NCX1 p omo es muscula dys ophy in mice. Mol. Cell. Biol. 34, 1991–2002
(2014).
24. Mackle , J. M., D ummond, J. A., Loewen, C. A., Robinson, I. M. & Reis , N. E.
The C(2)B Ca(2 þ)-binding mo i o synap o agmin is equi ed o synap ic
ansmission in i o.Na u e 418, 340–344 (2002).
25. Dale, J. M. e al. The spinal muscula a ophy mouse model, SMAD7, displays
al e ed axonal anspo wi hou global neu ofilamen al e a ions. Ac a
Neu opa hol. 122, 331–341 (2011).
26. Junie , M.-P. Wha ole(s) o TGFain he cen al ne ous sys em? P og.
Neu obiol. 62, 443–473 (2000).
27. Liso oski, F. e al. T ans o ming g ow h ac o alpha exp ession as a esponse
o mu ine mo o neu ons o axonal inju y and mu a ion-induced degene a ion.
J. Neu opa hol. Exp. Neu ol. 56, 459–471 (1997).
28. Boille
´e, S., Cadusseau, J., Coulpie , M., G annec, G. & Junie , M. P.
T ans o ming g ow h ac o alpha: a p omo e o mo oneu on su i al o
po en ial biological ele ance. J. Neu osci. 21, 7079–7088 (2001).
29. Deb ay, F.-G. e al. LRPPRC mu a ions cause a pheno ypically dis inc o m
o Leigh synd ome wi h cy och ome c oxidase deficiency. J. Med. Gene . 48,
183–189 (2011).
30. S einbe g, S. J. e al. Pe oxisome biogenesis diso de s. Biochim. Biophys. Ac a
1763, 1733–1748 2006.
31. Koolen, D. A. e al. Mu a ions in he ch oma in modifie gene KANSL1
cause he 17q21.31 mic odele ion synd ome. Na . Gene . 44, 639–641
ð2012Þ:
32. Nalls, M. A. e al. La ge-scale me a-analysis o genome-wide associa ion da a
iden ifies six new isk loci o Pa kinson’s disease. Na . Gene . 46, 989–993
(2014).
33. Wain, L. V. e al. No el insigh s in o he gene ics o smoking beha iou ,
lung unc ion, and ch onic obs uc i e pulmona y disease (UK BiLEVE): a
gene ic associa ion s udy in UK Biobank. Lance . Respi . Med. 3, 769–781
(2015).
34. Boe ge , L. M., Handsake , R. E., Zody, M. C. & McCa oll, S. A. S uc u al
haplo ypes and ecen e olu ion o he human 17q21.31 egion. Na . Gene . 44,
881–885 (2012).
35. S einbe g, K. M. e al. S uc u al di e si y and A ican o igin o he 17q21.31
in e sion polymo phism. Na . Gene . 44, 872–880 (2012).
36. Finucane, H. K. e al. Pa i ioning he i abili y by unc ional anno a ion
using genome-wide associa ion summa y s a is ics. Na . Gene . 47, 1228–1235
(2015).
37. Moo ha, V. K. e al. Iden ifica ion o a gene causing human cy och ome c
oxidase deficiency by in eg a i e genomics. P oc. Na l Acad. Sci. USA 100,
605–610 (2003).
38. Julien, M. e al. Phospha e-dependen egula ion o MGP in os eoblas s: ole o
ERK1/2 and F a-1. J. Bone Mine . Res. 24, 1856–1868 (2009).
39. Ba bei a, A. e al. In eg a ing issue specific mechanisms in o GWAS summa y
esul s. P ep in a bioRxi h ps://doi.o g/10.1101/045260 (2016).
40. Zhu, Z. e al. In eg a ion o summa y da a om GWAS and eQTL
s udies p edic s complex ai gene a ge s. Na . Gene . 48, 481–487
ð2016Þ:
41. Wes a, H.-J. e al. Sys ema ic iden ifica ion o ans eQTLs as
pu a i e d i e s o known disease associa ions. Na . Gene . 45, 1238–1243
(2013).
42. Han, H. Q., Zhou, X., Mi ch, W. E. & Goldbe g, A. L. Myos a in/ac i in
pa hway an agonism: molecula basis and he apeu ic po en ial. In . J. Biochem.
Cell Biol. 45, 2333–2347 (2013).
43. Cohen, S., Na han, J. A. & Goldbe g, A. L. Muscle was ing in disease:
molecula mechanisms and p omising he apies. Na . Re . D ug Disco . 14,
58–74 (2015).
44. Liu, J. Z. e al. A e sa ile gene-based es o genome-wide associa ion s udies.
Am. J. Hum. Gene . 87, 139–145 (2010).
45. Lach-T ifilie , E. e al. An an ibody blocking ac i in ype II ecep o s induces
s ong skele al muscle hype ophy and p o ec s om a ophy. Mol. Cell. Biol.
34, 606–618 (2014).
46. Ama o, A. A. e al. T ea men o spo adic inclusion body myosi is wi h
bimag umab. Neu ology 83, 2239–2246 (2014).
47. Yang, N. e al. ACTN3 geno ype is associa ed wi h human eli e a hle ic
pe o mance. Am. J. Hum. Gene . 73, 627–631 (2003).
48. She field-Moo e, M. & U ban, R. J. An o e iew o he endoc inology o
skele al muscle. T ends Endoc inol. Me ab. 15, 110–115 (2004).
49. Ho s man, A. M., Dillon, E. L., U ban, R. J. & She field-Moo e, M. The ole
o and ogens and es ogens on heal hy aging and longe i y. J. Ge on ol. 67,
1140–1152 (2012).
50. Gulle , N. P., Hebba , G. & Ziegle , T. R. Upda e on clinical ials o g ow h
ac o s and anabolic s e oids in cachexia and was ing 1–4. Cu . Opin. Clin.
Nu . Me ab. Ca e 91, 1143–1147 (2010).
51. Lawlo , D. A., Ha bo d, R. M., S e ne, J. A. C., Timpson, N. & Da ey Smi h, G.
Mendelian andomiza ion: using genes as ins umen s o making causal
in e ences in epidemiology. S a . Med. 27, 1133–1163 (2008).
52. Bu gess, S., Bu e wo h, A. & Thompson, S. G. Mendelian andomiza ion
analysis wi h mul iple gene ic a ian s using summa ized da a. Gene .
Epidemiol. 37, 658–665 (2013).
53. Co iello, A. D. e al. A genome-wide associa ion me a-analysis o ci cula ing
sex ho mone-binding globulin e eals mul iple Loci implica ed in sex s e oid
ho mone egula ion. PLoS Gene . 8, e1002805 (2012).
54. Zhai, G. e al. Eigh common gene ic a ian s associa ed wi h se um
DHEAS le els sugges a key ole in ageing mechanisms. PLoS Gene . 7,
e1002025 (2011).
55. Teume , A. e al. Genomewide me a-analysis iden ifies loci associa ed wi h
IGF-I and IGFBP-3 le els wi h impac on age- ela ed ai s. Aging Cell 15,
811–824 (2016).
56. Joshi, P. K. e al. Va ian s nea CHRNA3/5 and APOE ha e age- and
sex- ela ed e ec s on human li espan. Na . Commun. 7, 11174 (2016).
57. Nikpay, M. e al. A comp ehensi e 1,000 Genomes-based genome-wide
associa ion me a-analysis o co ona y a e y disease. Na . Gene . 47, 1121–1130
(2015).
58. Zheng, H. e al. Whole-genome sequencing iden ifies EN1 as a de e minan o
bone densi y and ac u e. Na u e 526, 112–117 (2015).
59. Clemson, L. e al. In eg a ion o balance and s eng h aining in o daily li e
ac i i y o educe a e o alls in olde people ( he LiFE s udy): andomised
pa allel ial. BMJ 345, e4547 (2012).
60. Edwa ds, B. J., Song, J., Dunlop, D. D., Fink, H. A. & Cauley, J. A. Func ional
decline a e inciden w is ac u es--S udy o Os eopo o ic F ac u es:
p ospec i e coho s udy. BMJ 341, c3324–c3324 (2010).
61. Huang, J. e al. Imp o ed impu a ion o low- equency and a e
a ian s using he UK10K haplo ype e e ence panel. Na . Commun. 6, 8111
(2015).
62. Loh, P.-R. e al. E ficien Bayesian mixed-model analysis inc eases associa ion
powe in la ge coho s. Na . Gene . 47, 284–290 (2015).
63. Bulik-Sulli an, B. K. e al. LD Sco e eg ession dis inguishes con ounding om
polygenici y in genome-wide associa ion s udies. Na . Gene . 47, 291–295
(2015).
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16015 ARTICLE
NATURE COMMUNICATIONS | 8:16015 | DOI: 10.1038/ncomms16015 | www.na u e.com/na u ecommunica ions 9