RESEARCH ARTICLE
MiRNA P o iles in Lymphoblas oid Cell Lines
o Finnish P os a e Cance Families
Daniel Fische
1
, Tiina Wahl o s
2
, Henna Ma ila
2
, Hannu Oja
3
, Teu o L. J. Tammela
4
,
Johanna Schleu ke
5
*
1School o Heal h Sciences, Uni e si y o Tampe e, 33014 Tampe e, Finland, 2BioMediTech, Uni e si y o
Tampe e, and Fimlab Labo a o ies, Tampe e, Finland, 3Depa men o Ma hema ics and S a is ics,
Uni e si y o Tu ku, 20014 Tu ku, Finland, 4Depa men o U ology, Tampe e Uni e si y Hospi al and
Medical School, Uni e si y o Tampe e, Tampe e, Finland, 5Medical Biochemis y and Gene ics, Ins i u e o
Biomedicine, Uni e si y o Tu ku, Tu ku, Finland
*johanna.schleu ke @u u. i
Abs ac
Backg ound
He i able ac o s a e e iden ly in ol ed in p os a e cance (P Ca) ca cinogenesis, bu cu -
en ly, gene ic ma ke s a e no ou inely used in sc eening o diagnos ics o he disease.
Mo e p ecise in o ma ion is needed o making ea men decisions o dis inguish agg es-
si e cases om indolen disease, o which he i able ac o s could be a use ul ool. The ge-
ne ic makeup o P Ca has only ecen ly begun o be un a elled h ough la ge-scale
genome-wide associa ion s udies (GWAS). The hus a iden i ied Single Nucleo ide Poly-
mo phisms (SNPs) explain, howe e , only a ac ion o amilial clus e ing. Mo eo e , he
known isk SNPs a e no associa ed wi h he clinical ou come o he disease, such as ag-
g essi e o me as asised disease, and he e o e canno be used o p edic he p ognosis.
Anno a ing he SNPs wi h deep clinical da a oge he wi h miRNA exp ession p o iles can
imp o e he unde s anding o he unde lying mechanisms o di e en pheno ypes o
p os a e cance .
Resul s
In his s udy mic oRNA (miRNA) p o iles we e s udied as po en ial bioma ke s o p edic he
disease ou come. The s udy subjec s we e om Finnish high isk p os a e cance amilies.
To iden i y po en ial bioma ke s we combined a no el non-pa ame ical es wi h an impo -
ance measu e p o ided om a Random Fo es classi ie . This combina ion deli e ed a se
o nine miRNAs ha was able o sepa a e cases om con ols. The de ec ed miRNA ex-
p ession p o iles could p edic he de elopmen o he disease yea s be o e he ac ual P Ca
diagnosis o de ec he exis ence o o he cance s in he s udied indi iduals. Fu he mo e,
using an exp ession Quan i a i e T ai Loci (eQTL) analysis, egula o y SNPs o miRNA
miR-483-3p ha we e also di ec ly associa ed wi h P Ca we e ound.
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 1/17
a11111
OPEN ACCESS
Ci a ion: Fische D, Wahl o s T, Ma ila H, Oja H,
Tammela TLJ, Schleu ke J (2015) MiRNA P o iles in
Lymphoblas oid Cell Lines o Finnish P os a e Cance
Families. PLoS ONE 10(5): e0127427. doi:10.1371/
jou nal.pone.0127427
Academic Edi o : Xin-Yuan Guan, The Uni e si y o
Hong Kong, CHINA
Recei ed: Decembe 19, 2014
Accep ed: Ap il 15, 2015
Published: May 28, 2015
Copy igh : © 2015 Fische e al. This is an open
access a icle dis ibu ed unde he e ms o he
C ea i e Commons A ibu ion License, which pe mi s
un es ic ed use, dis ibu ion, and ep oduc ion in any
medium, p o ided he o iginal au ho and sou ce a e
c edi ed.
Da a A ailabili y S a emen : All ele an da a a e
a ailable om EBI (accession numbe E-MTAB-
3397).
Funding: This wo k was suppo ed by Medical
Resea ch Fund o Tampe e Uni e si y Hospi al
(9L091, 9M094, and 9N069), he Finnish Cance
O ganiza ions, he Sig id Juselius Founda ion, and
he Academy o Finland (g an s 116437 and 251074)
o JS. This wo k was also suppo ed by The Finnish
Doc o al P og amme in S ochas ics and S a is ics o
DF.
Conclusion
Based on ou indings, we sugges ha blood-based miRNA exp ession p o iling can be
used in he diagnosis and maybe e en p ognosis o he disease. In he u u e, miRNA p o il-
ing could possibly be used in a ge ed sc eening, oge he wi h P os a e Speci ic An igene
(PSA) es ing, o iden i y men wi h an ele a ed P Ca isk.
In oduc ion
P os a e cance (P Ca) is he mos common noncu aneous malignancy and he second leading
cause o cance - ela ed dea hs among men in indus ialised coun ies [1]. In Finland, 4604 new
p os a e cance cases we e diagnosed in 2012 (Finnish Cance Regis y, h p://www.cance . i/
syopa ekis e i/). Aging and PSA es ing may be he mos e iden easons o he inc eased
numbe o new cases. The g owing incidence c ea es p essu e on he heal h ca e sys em as he
conce n ega ding o e ea men is conside able. The e o e, one o he majo challenges is o
imp o e he diagnos ic and p ognos ic ools o be able o dis inguish le hal om indolen dis-
ease a a cu able s a e o he disease.
The con ibu ion o gene ic a ian s has been s udied widely in associa ion wi h p os a e
cance p edisposi ion. Bo h linkage and GWAS oge he wi h he ew examples a ising om
candida e gene app oaches ha e led o he iden i ica ion o abou 100 gene ic loci ha explain
only app oxima ely 30% o he gene ic isk o he disease [2][3][4][5]. Howe e , he e is no
ob ious molecula o unc ional e idence indica ing how he a ia ions in hese candida e si es
o hei co-inhe i ed neighbou ing a ian s could cause P Ca. In ac , mos o he single nucleo-
ide a ian s (SNPs) ound by GWAS a e unlikely o a ec he coding sequence o any gene bu
a he eside in in e genic egions. These indings sugges ha hey ha e a egula o y ole, such
as in ansc ip ion, splicing o mRNA s abili y, ins ead o a di ec e ec on he unc ion o he
gene p oduc [6].
In ecen yea s, he impo ance o he non-p o ein coding genome in he unc ional egula-
ion o no mal de elopmen and disease de elopmen has become e iden . MiRNAs a e sho
non-coding RNAs ha egula e hei a ge gene exp ession ypically by binding o he 3’un-
ansla ed egion (UTR) o he a ge mRNA [7]. Indi idual a ia ion o he miRNA exp ession
le els can in luence he exp ession o he mRNA a ge gene, causing pheno ypic di e ences.
Se e al s udies ha e shown ha miRNA exp ession le els a e p edic i e o he ou come o
solid umou s and leukaemias, bu he con ibu ion o al e ed miRNA exp ession le els o ge-
ne ic cance suscep ibili y is no known. The ansc ip ional ac i i y o p o ein coding genes is
inhe i ed as a quan i a i e ai , and egula o y polymo phisms associa ed wi h he a iabili y
in he le els o mRNA a e conside ed o be eQTL. Despi e he demons a ed impo ance,
knowledge o he gene ic egula ion o miRNA exp ession is s ill in i s in ancy. In a ecen pub-
lica ion, o e one hund ed eQTLs in p ima y ib oblas s we e desc ibed, indica ing a leas a
pa ial ole o gene ic a ia ion in al e ed miRNA exp ession [8]. Combined analyses o com-
mon SNPs and a ia ions in miRNA exp ession p o iles migh se e as one way o elucida e
he biological unc ions o SNPs iden i ied om GWAS in common diseases.
The objec i e o his s udy was o e alua e he miRNA exp ession p o iles o lymphoblas oid
cell lines (LCL) de i ed om membe s o high isk P Ca amilies. Al e ed miRNA exp ession
in pa ien LCLs compa ed wi h hose om heal hy amily membe s p o ided an oppo uni y
o iden i y ge mline a ian s in p omo e o o he egula o y egions o p o ein coding genes as
a conside able amoun o miRNA exp ession is co ela ed o hos and a ge gene exp ession
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 2/17
Compe ing In e es s: The au ho s ha e decla ed
ha no compe ing in e es s exis .
[9]. The la ge amoun o signi ican miRNA-wise es esul s wi hin he da a also equi ed he
de elopmen o a new ype o di e en ially exp ession analysis pipeline. To de elop such a
pipeline, di e en ially exp ession es ing has been combined wi h he impo ance measu es o
he machine lea ning algo i hm, Random Fo es [10].
Ma e ials and Me hods
E hics S a emen
This s udy has been app o ed by he espec i e IRB boa ds o The Minis y o Social A ai s
and Heal h (SMT), Na ional Supe iso y Au ho i y o Wel a e and Heal h (Val i a) and E h-
ics Commi ee o Tampe e Uni e si y Hospi al. E e y indi idual pa icipa ing in he s udy has
gi en w i en in o med consen .
S udy popula ion
All samples a e o Finnish o igin and he collec ion o he amilies has been epo ed p e iously
[11]. Fo he miRNA mic oa ay s udy, 115 cases om 70 P Ca amilies we e used. The selec -
ed amilies had a leas wo i s -deg ee ela i es diagnosed wi h p os a e cance a any age.
Heal hy (= no diagnosed p os a e cance ) indi iduals (n = 78) om 47 amilies we e used as
he con ols. The median age a diagnosis o he cases was 65 (44–86.2) yea s and he con ols
had a median age o 57.5 (35.2–83.3) yea s a he ime he samples we e ob ained.
A subse o indi iduals (n = 54) om he mic oa ay expe imen we e geno yped wi h Illu-
mina’s HumanOmniExp ess a ay o ano he expe imen , and he esul s a e published else-
whe e [12]. Hence, hose 54 samples could be used he e o an eQTL analysis (39 P Ca cases
and 15 con ols). Addi ional 83 indi iduals could be used o alida ion pu poses. Al oge he ,
he e we e 137 geno yped pe sons om 33 amilies (20 o e lapping amilies wi h he mic oa -
ay pa o he s udy).
The clinical ou come o p os a e cance can oughly be classi ied in o agg essi e and non-
agg essi e cance , based on PSA, Gleason sco e and o he clinical e alua ions [13]. Based on
hese guidelines, he p os a e cance pa ien s om he wo expe imen s we e g ouped in o 36
(36) agg essi e and 79 (66) non-agg essi e p os a e cance s. The maximum numbe o agg es-
si e cases pe amily was 3, and he minimum was 1. A de ailed o e iew o he indi iduals in
he s udy is gi en in Fig 1.
RNA ex ac ion om lymphoblas oid cell lines
LCLs we e de i ed by he Eps ein-Ba i us ans o ma ion o pe iphe al mononuclea leuko-
cy es om pa ien s and hei heal hy ela i es. The lymphoblas oid cell lines we e g own in
RPMI-1640 medium (Lonza, Walke s ille, MD, USA) supplemen ed wi h 10% e al bo ine
se um (Sigma-Ald ich, S . Louis, MO, USA) and an ibio ics a 37°C, 5% CO2 and 95% hu-
midi y. The cell pelle s we e snap- ozen, and o al RNA was ex ac ed wi h T izol acco ding o
he manu ac u e ’s ins uc ions (In i ogen, Ca lsbad, CA, USA). The RNA yields we e quan i-
ied using an ND-1000 spec opho ome e (Nanod op Technologies, Wilming on, DE, USA)
and Agilen 2100 Bioanalyze (Agilen Technologies, San a Cla a, CA, USA).
Mic oRNA mic oa ay analysis
The mic oRNA exp ession le els in LCLs we e de ec ed using Agilen Human miRNA V2
Oligo Mic oa ay Ki (Agilen Technologies). Fi s , 100 ng o o al RNA was used as he s a -
ing ma e ial, and miRNAs we e labeled using he Agilen miRNA Labelling Ki . Labelled RNA
was hyb idised o Agilen miRNA mic oa ays ha ha e eigh iden ical a ays pe slide, wi h
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 3/17
each a ay con aining p obes di ec ed agains 817 miRNAs (719 human, 76 non-human i al
miRNAs and 22 con ol miRNAs). In o al, 26 slides we e used, and he da a we e ex ac ed
using Agilen ’s Fea u e Ex ac ion so wa e (FES), e sion 10.7.1.1 wi h he g id layou
D_F_20091030. Fo he da a analysis, low quali y samples we e i s emo ed, esul ing in 193
indi iduals. Each indi idual Agilen mic oa ay V2 measu es 13,737 ea u es, and he FES hen
used hese ea u es o calcula e he exp ession alues o 2,466 (2,125 human) p obes; based on
hose p obes he 817 miRNA exp ession alues we e calcula ed. The da a can be accessed ia
A ayExp ess accession E-MTAB-3397.
The miRNA exp ession alues a e ypically calcula ed wi h he algo i hm gTo alGeneSignal
as implemen ed in FES, bu in his s udy, howe e , p obe-wise, backg ound sub ac ed median
alues we e used ins ead. The analysis o di e en p obes o he same miRNA as a single
miRNA exp ession alue did no appea o be eliable enough, and an analysis a he p obe
le el was mo e easible. A e calcula ing he exp ession alues a he p obe le el, all non-
human p obes and hose no de ec ed by he FES we e emo ed. Only hose p obes ha we e
de ec ed o a leas 50% o he samples in a leas one heal h s a us g oup we e used o u he
analysis. Addi ionally, non-human con ol ea u es we e emo ed be o e he analysis. In o al,
547 p obes, ep esen ing 211 miRNAs, ul illed hese c i e ia. The echnical a iabili y o he
da a was educed by applying a quan ile no malisa ion [14].
Geno yping Da a Analysis
The single nucleo ide polymo phism (SNP) geno ype da a we e gene a ed using Illumina’s
HumanOmniExp ess a ay in collabo a ion wi h he Ins i u e o Molecula Medicine Finland
(FIMM). The chosen a ay enabled he geno yping o app oxima ely 700k SNPs. To p oduce
he geno ype da a, he aw da a we e analysed wi h Genome S udio acco ding o he manu ac-
u e ’s ins uc ions (Illumina, San Diego, USA).
In o al, he geno ype in o ma ion o 137 indi iduals was a ailable, wi h he miRNA ex-
p ession le els also measu ed in 54 o hese indi iduals. Hence, he eQTL analysis was based
on hese 54 pe sons. The emaining 83 indi iduals we e used o alida ion o he esul s.
Fig 1. uppe : Popula ion quan i ies, isualisa ion o how he 277 indi iduals in his s udy a e dis ibu ed among he h ee heal h-s a us g oups. Fo
each heal h g oup, he numbe o indi iduals om he di e en expe imen s is shown. The o e all numbe om an expe imen is hen indica ed by he
espec i e colou ed box plus he ed box (o e lap). lowe : Visualisa ion o he amilial backg ound. The h ee op ions ‘P Ca only’,‘Heal hy only’o ‘P Ca/
Heal hy’a e shown and g ouped acco dingly. Addi ionally, in ol emen o di e en amilies in he wo expe imen s is shown. O de ing is acco ding o an
in e nal amily code.
doi:10.1371/jou nal.pone.0127427.g001
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 4/17
Iden i ica ion o di e en ially exp essed p obes using di ec ional es ing
P Ca pa ien s we e di ided in o agg essi e (A) and non-agg essi e/mild (M) P Ca g oups and
compa ed wi h heal hy con ols (H). A new gene alisa ion o Mann-Whi ney ype es s was ap-
plied o iden i y di e en ially exp essed p obes in he h ee-g oup compa ison. The same gen-
e alisa ion was used o he eQTL analysis ( o de ails see [15] and [16]).
Fo a gene al de ini ion, le he sample sizes o he h ee g oups be N
H
,N
M
and N
A
which e-
sul s in a o al sample size o N
H
+N
M
+N
A
=N. The gene alised Mann-Whi ney es is based
on p obabilis ic indices calcula ed wi h iple sums o co esponding indica o unc ions. Le
x
p;H
=(x
1,p;H
,x
2,p;H
,...,x
N
H
,p;H
)
T
,x
p;M
=(x
1,p;M
,x
2,p;M
,...,x
N
M
,p;M
)
T
and x
p;A
=(x
1,p;A
,x
2,p;A
,
...,x
N
A
,p;A
)
T
be he exp ession alues o a p obe pin each heal h g oup wi h unde lying cd ’s
F
p;H
,F
p;M
and F
p;A
. The p obabilis ic index ^
PH;M;A;p o p obe pused in his app oach can hen
be calcula ed by
^
PH;M;A;p¼1
NHNMNA
X
NH
i¼1
X
NM
j¼1
X
NA
k¼1
Iðxi;p;H<xj;p;M<xk;p;AÞ;
and I() is he indica o unc ion ha is 1 i condi ion () is ue and 0 i no . Please no ice ha
he o de in he index o ^
PH;M;A;p e e s o he o de used in he indica o unc ion.
Fu he mo e, he p obabilis ic index ^
PH;M;A;pcan hen be used o es he di ec ional hypo h-
esis
H0:Fp;H¼Fp;M¼Fp;A s:H1:Fp;HFp;MFp;A;
whe e  e e s o he s ochas ic o de ing o cd ’s. Na u ally, di e en o de s in he condi ion
() o he indica o unc ion can be used o es o di e en al e na i es. In addi ion, when ex-
p ession alues a e assigned o geno ype g oups ins ead o heal h s a us, his es p ocedu e is
ideal o eQTL es ing as i es s o he di ec ional al e na i es ha a e clea ly p esen in he
con ex o an eQTL analysis.
The wo p obabilis ic indices ^
PH;M;A;pand ^
PA;M;H;pwe e used o es ing p obes p=1,...,
547, and p- alues o he pe mu a ion es e sion we e calcula ed based on 5000 pe mu a ions.
Tes esul s wi h p- alue less han 0.01 we e conside ed o be signi ican . The es me hod is
implemen ed in he R-package gMWT [16], and he package Gene icTools exploi s his es
me hod o eQTL es ing. Bo h packages a e eely a ailable om he Comp ehensi e R A -
chi e Ne wo k (CRAN).
The Benjamini-Hochbe g mul iple es ing p ocedu e o con ol he alse disco e y a e is i-
sualised using ejec ion plo s and lines. The a io o expec ed ejec ions unde he null hypo he-
sis is plo ed agains he obse ed a io o ejec ions. I his cu e is abo e he (0, 1)-line, we
ha e mo e ejec ions han expec ed unde he null hypo hesis. The ejec ions o a ixed es
size can be isualised wi h a e ical line, and he ejec ions o di e en mul iple es ing adjus -
men s can be isualised by lines wi h a ce ain slope. The numbe o ejec ed null hypo heses is
hen de e mined by he c ossing poin o he cu e and he line. Fo de ails, see [15].
Classi ica ion, Impo ance Measu e and Clus e ing
The machine lea ning classi ie Random Fo es [10], as implemen ed in he R-package an-
domFo es [17], was applied o he exp ession da a, such ha he da ase was spli in o he
aining (75%) and es (25%) da a. The aining da a we e used o c ea e an ensemble o 2500
decision ees, and hese ees we e hen used o classi y he es da a. The di ision be ween he
aining and alida ion da a was hen epea ed 2000 imes, and a e wa ds he classi ica ion
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 5/17

esul s o all es da a uns we e e alua ed. The Gini impo ance measu e was also ex ac ed o
e e y single Random Fo es , and he a e age impo ance o each p obe was combined wi h he
co esponding p- alue om he di ec ional es . P obes ha had a p- alue less han 0.01 and
ha belonged o he 10% mos impo an p obes o e all Random Fo es uns we e conside ed
o be o high in e es (HI p obes) and we e hen used in he clus e ing s ep and in he
eQTL analysis.
The Random Fo es s we e ained o he h ee possible ou come classes heal hy (H), mild
P Ca (M) and agg essi e P Ca (A). Le L
i, ;H
,L
i, ;M
and L
i, ;A
be he class likelihoods p o ided
by he Random Fo es classi ie un o indi idual iwi h L
i, ;H
+L
i, ;M
+L
i, ;A
= 1. These likeli-
hoods we e hen combined in o a single P Ca se e eness alue Si; ¼1
2Li; ;MþLi; ;A. The se e -
ness alue S
i,
was chosen in such a way ha S
i,
= 0 in case ha L
i, ;H
=1,S
i,
= 0.5 o L
i, ;M
=1
and S
i,
=1i L
i, ;A
=1.
In a 2-way Random Fo es un, he classi ica ion was pe o med only be ween he heal hy
and P Ca classes, wi h same se up as ha o he 3-way Random Fo es desc ibed abo e.
To calcula e he A ea Unde he Cu e (AUC) o he Recei e Ope a ing Cha ac e is ic
(ROC) cu e in he Random Fo es case, wo di e en app oaches we e chosen. Fi s , he wo
likelihoods L
i, ;M
and L
i, ;A
we e added o e alua e he Random Fo es ’s capabili y o classi y
P Ca in gene al. Then, in he second compa ison, he likelihoods L
i, ;H
and L
i, ;M
we e added o
e alua e i s ap i ude o iden i y agg essi e P Ca. E en ually, o plo he ROC a con inuous cu -
o alue in [0, 1] was applied on o he likelihood o classi y indi iduals in o ue/
alse posi i es.
Fo he clus e ing in he hea map, he Kendall au co ela ion ma ix Samong all samples
was calcula ed based on he exp ession alues o he HI p obes. Kendall’ au be ween wo a i-
ables is a measu e o posi i e/nega i e dependence and is in a ian unde any s ic ly inc eas-
ing ans o ma ion o he ma ginal a iables. The co esponding dis ance be ween he
a iables is hen de ined as D=(1−S)/2. Le hen Dbe he ma ix o dis ances used o he
hie achical clus e ing.
eQTL Analysis
The geno ype in o ma ion om he 700k a ay was combined wi h he exp ession alues o he
HI p obes using an eQTL analysis. The ch omosomal loca ions o he miRNA p obes we e
iden i ied and all SNPs wi hin a window o 1Mb a ound he p obe’s cen al loca ion we e
linked o his p obe. The p obe exp ession alues we e hen assigned o he geno ype g oups o
e e y linked SNP (Fig 2 shows a sys ema ic ske ch o his s ep).
In an eQTL app oach, h ee cases a e possible, depending on whe he he exp ession alues
ha e been assigned o one, wo o all h ee possible geno ype g oups. Monomo phic a ian s
we e no u he conside ed in he analysis, and in he wo-g oup case, a wo-sided Mann-
Whi ney es was applied. In he h ee-g oup case, he gene alised Mann-Whi ney es o di-
ec ional al e na i es was used o he wo di e en al e na i es whe he he highe exp ession
alues we e linked o he wild- ype o he homozygous mu a ion. This ype o di ec ional es
was used in he h ee-g oup case as an o de o he exp ession alues wi h espec o he geno-
ype g oups is clea ly expec ed.
Compa a i e Analysis
The he e used wo-s age app oach was compa ed wi h wo o he commonly used me hods.
The i s me hod was a classical Analysis o Va iance (ANOVA), es ing he al e na i e hypo h-
esis ha he e is a di e ence be ween a leas wo ou o he h ee g oups. Le μ
p,H
,μ
p,M
and μ
p,A
be he a e age exp ession alues o p obe p o he h ee g oups, hen is he p obe-wise
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 6/17
hypo hesis o he one-way ANOVA
H0:mp;H¼mp;M¼mp;A s:H1:No all mp;: a e equal
Resul ing p- alues we e hen adjus ed o mul iple es ing using a bon e oni co ec ion.
The second me hod ha was used as compa ison was a wo-s aged logis ic eg ession wi h
lasso (LRL). Fi s , LRL was applied on o he ull da ase wi h he wo classes heal hy/diseased.
The uning pa ame e λwas chosen such ha he amoun o selec ed a iables we e in he
same le el o magni ude as he he e p oposed me hod iden i ies. The second LRL un was hen
applied on o he cance cases only and aimed o he sepa a ion o mild and agg essi e P Ca.
Finally he esul ing p obes we e me ged o one esul ma ix om he LRL analysis.
To compa e he esul s o he ANOVA and he LRL wi h he he e p oposed app oach, a hi-
e a chical clus e ing was applied on o he iden i ied p obes using also a Kendall’s au based dis-
ance ma ix. Then, he adjus ed Rand Index was calcula ed be ween he classi ica ion o he
h ee di e en clus e ings and he ue cance s a us o he indi iduals o de e mine he le el
o ag eemen .
Resul s
Using he di ec ional es ing p ocedu e, 146 (87 wi h highe exp ession in agg essi e P Ca and
59 wi h highe exp ession in con ols) ou o a o al o 547 p obes we e iden i ied ha ing di e -
en exp ession p o iles. The ch omosomal loca ion o he signi ican p obes and he ype o
es ing al e na i e a e isualised in Fig 3.
To iden i y HI p obes om his unexpec edly la ge amoun o di e en ially exp essed
p obes, a Random Fo es classi ie was also applied o he exp ession da a. Signi ican p obes
ha we e wi hin 10% o he mos impo an p obes in he Random Fo es , measu ed as Gini
Index, we e called HI p obes and a e highligh ed in Fig 3. The 13 iden i ied p obes ep esen
eigh di e en miRNAs and one spliceosomal RNA. Mo e de ails abou he 13 iden i ied p obes
a e lis ed in Table 1.
The o e all classi ica ion esul based on he se e eness alues S
i,
o he Random Fo es is
isualised in Fig 4. Heal hy indi iduals (g een) clea ly ended o be in he lowe isk a ea, bu
Fig 2. Each line ep esen s an indi idual, ha ing a ce ain exp ession alue o miRNA X. Independen o he heal h s a us o each indi idual, he
exp ession alues a e g ouped acco ding o he geno ype g oups o he su ounding SNPs and hen es ed o di e en ial exp ession be ween hose g oups.
(Figu e aken om [16])
doi:10.1371/jou nal.pone.0127427.g002
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 7/17
agg essi e P Ca pa ien s ( ed) did no end o ha e la ge alues han non-agg essi e P Ca pa-
ien s (yellow). In addi ion, an a e age classi ica ion a e o e all classi ica ion uns was de e -
mined sepa a ely o he compa isons be ween heal hy and P Ca and be ween agg essi e P Ca
and combined heal hy and non-agg essi e P Ca. The Random Fo es was able o classi y P Ca
wi h an a e age AUC o he ROC o app oxima ely 0.89 and agg essi e P Ca e sus he com-
bined samples o non-agg essi e P Ca and con ols o 0.68 (Fig 5). The classi ica ion esul s a
he indi idual le el a e isualised in he suppo ing in o ma ion (S1 and S2 Figs).
A hie a chical clus e ing shows he impo ance o he HI p obes. Clus e ing he da ase
based on all p obes esul ed in only a sligh ly be e classi ica ion han he clus e ing based on
Fig 3. Loca ion o he di ec ional es esul s o he wo p obabilis ic indicies ^
PH;M;Aand ^
PA;M;Hdeno ed by H<M<A espec i e A<M<H.
Signi ican es esul s ha also belong o he 10% mos impo an (Gini Index) miRNAs in he Random Fo es un a e deno ed as HI p obes.
doi:10.1371/jou nal.pone.0127427.g003
Table 1. O e iew o he HI P obes, hei a ge miRNAs wi h co esponding median exp ession alues and ch omosomal posi ion.
P obeID Ta ge ID Ch omosomalLoca ion ~
xH~
xM~
xA
A_25_P00010263 mi |hsa-miR-328 Ch 16:67,236,292—67,236,276 20.81 27.15 25.75
A_25_P00011068 mi |hsa-miR-107 Ch 10:91,352,575—91,352,557 405.20 483.35 483.35
A_25_P00011440 mi |hsa-miR-801_ 10.1 Ch 1:28,847,749—28,847,763 50.77 41.10 34.29
A_25_P00011476 mi |hsa-miR-770-5p Ch 14:101,318,754—101,318,768 28.59 24.52 24.13
A_25_P00011477 mi |hsa-miR-770-5p Ch 14:101,318,755—101,318,768 24.59 20.52 20.53
A_25_P00011979 mi |hsa-miR-770-5p Ch 14:101,318,752—101,318,768 31.16 26.31 26.08
A_25_P00012461 mi |hsa-miR-483-3p Ch 11:2,155,431—2,155,415 24.72 29.56 29.94
A_25_P00012462 mi |hsa-miR-483-3p Ch 11:2,155,431—2,155,414 23.46 29.28 29.35
A_25_P00012991 mi |hsa-miR-885-5p Ch 3:10,436,204—10,436,189 22.43 33.34 32.14
A_25_P00013086 mi |hsa-miR-939 Ch 8:145,619,401—145,619,390 152.93 71.98 72.08
A_25_P00013207 mi |hsa-miR-29a*Ch 7:13,0561,530—130,561,511 18.89 21.36 21.52
A_25_P00014864 mi |hsa-miR-202 Ch 10:135,061,097—135,061,083 25.92 21.29 20.68
A_25_P00014914 mi |hsa-miR-885-5p Ch 3:10,436,204—10,436,188 21.25 30.92 29.61
doi:10.1371/jou nal.pone.0127427. 001
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 8/17
he 13 HI p obes. The dend og am o clus e ing indi iduals based on he 13 HI p obes oge h-
e wi h he co esponding hea map is shown in Fig 6. He e, he abili y o sepa a e clea ly be-
ween agg essi e and non-agg essi e P Ca was limi ed, bu in e es ingly only i e o he 78
heal hy indi iduals we e clus e ed closely oge he wi h P Ca indi iduals. In con as , 46 o 115
P Ca cases we e inside he clus e ha con ained mos o he heal hy indi iduals.
In addi ion, a cis-eQTL (0.5Mb up/downs eam window) o he HI p obes was pe o med.
In o al, 3863 SNP-miRNA associa ions we e es ed, and 79 had a p- alue o 0.01, (S3 Fig in
he suppo ing in o ma ion). All SNPs ha we e ound o ha e a possible egula o y e ec on
an HI p obe we e hen es ed o a di ec P Ca associa ion by applying a Fishe - es on he
2 × 3 able be ween geno ype and heal h s a us g oups. Fo ou SNPs, a signi ican associa ion
was ound o he 53 geno ypes o he eQTL samples ( es size 0.05).
Fig 4. O e all classi ica ion esul s o he Random Fo es classi ie using he se e eness measu e S
i,
.
doi:10.1371/jou nal.pone.0127427.g004
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 9/17
Re e ences
1. And iole GL, C aw o d ED, G ubb RL, Buys SS, Chia D, Chu ch TR, e al. Mo ali y esul s om a an-
domized p os a e-cance sc eening ial. N Engl J Med. 2009; 360(13): 1310–1319. doi: 10.1056/
NEJMoa0810696 PMID: 19297565
2. Va ghese JS, Eas on DF. Genome-wide associa ion s udies in common cance s—wha ha e we
lea n ? Cu Opin Gene De . 2010; 20(3): 201–209. PMID: 20418093
3. Eeles RA, Olama AA, Benlloch S, Saunde s EJ, Leongamo nle DA, Tym akiewicz M, e al. Iden i ica-
ion o 23 new p os a e cance suscep ibili y loci using he iCOGS cus om geno yping a ay. Na Gene .
2013; 45(4): 385–391. doi: 10.1038/ng.2560 PMID: 23535732
4. Kim ST, Cheng Y, Hsu FC, Jin T, Kade AK, Zheng SL, e al. P os a e cance isk-associa ed a ian s
epo ed om genome-wide associa ion s udies: Me a-analysis and hei con ibu ion o gene ic a ia-
ion. P os a e. 2010; 70(16): 1729–1738. doi: 10.1002/p os.21208 PMID: 20564319
5. So HC, Gui AH, Che ny SS, Sham PC. E alua ing he he i abili y explained by known suscep ibili y a -
ian s: A su ey o en complex diseases. Gene Epidemiol. 2011; 35(5): 310–317. doi: 10.1002/gepi.
20579 PMID: 21374718
6. Nicolae DL, Gamazon E, Zhang W, Duan S, Dolan ME, Cox NJ. T ai -Associa ed SNPs A e Mo e Likely
o Be eQTLs: Anno a ion o Enhance Disco e y om GWAS. PLoS Gene . 2010; 6(4):e1000888. doi:
10.1371/jou nal.pgen.1000888 PMID: 20369019
7. Busha i N, Cohen SM. Mic oRNA unc ions. Annu Re Cell De Biol. 2007; 23: 175–205. doi: 10.1146/
annu e .cellbio.23.090506.123406 PMID: 17506695
8. Bo el C, Deu sch S, Le ou neau A, Miglia acca E, Mon gome y SB, Dimas AS, e al. Iden i ica ion o
cis- and ans- egula o y a ia ion modula ing mic oRNA exp ession le els in human ib oblas s. Ge-
nome Res. 2011; 21(1): 68–73. doi: 10.1101/g .109371.110 PMID: 21147911
9. Lu e D, Ma C, K umsiek J, Lang EW, Theis FJ. In onic mic oRNAs suppo hei hos genes by medi-
a ing syne gis ic and an agonis ic egula o y e ec s. BMC Genomics. 2010; 11(224): 11–224.
10. B eiman L. Random o es s. Mach Lea n. 2001;p. 5–32. doi: 10.1023/A:1010933404324
11. Schleu ke J, Ma ikainen M, Smi h J, Koi is o P, Ba oe-Bonnie A, Kainu T, e al. A gene ic epidemiolog-
ical s udy o he edi a y p os a e cance (HPD) in Finland: equen HPCX linkage in amilies wi h la e-
onse disease. Clin Cance Res. 2000; 6(12): 4810–4815. PMID: 11156239
12. Sil anen S, Fische D, Ran ape o T, Lai inen V, Mpindi JP, Kallioniemi O, e al. ARLTS1 and P os a e
Cance Risk—Analysis o Exp ession and Regula ion. PLoS One. 2013; 8(8):e72040. doi: 10.1371/
jou nal.pone.0072040 PMID: 23940804
13. Schaid DJ, McDonnell SK, Za as KE, Cunningham JM, Hebb ing S, Thibodeau SN, e al. Pooled ge-
nome linkage scan o agg essi e p os a e cance : esul s om he In e na ional Conso ium o P os a e
Cance Gene ics. Hum Gene . 2006; 120(4): 471–485. doi: 10.1007/s00439-006-0219-9 PMID:
16932970
14. Bols ad BM, I iza y RA, As and M, Speed TP. A compa ison o no maliza ion me hods o high densi y
oligonucleo ide a ay da a based on a iance and bias. Bioin o ma ics. 2003; 19(2): 185–193. doi: 10.
1093/bioin o ma ics/19.2.185 PMID: 12538238
15. Fische D, Oja H, Schleu ke J, Sen PK, Wahl o s T. Gene alized Mann-Whi ney Type Tes s o Mic o-
a ay Expe imen s. Scand S a Theo y Appl. 2014; 41(3): 672–692. doi: 10.1111/sjos.12055
16. Fische D, Oja H. Mann-Whi ney Type Tes s o Mic oa ay Expe imen s: The R Package gMWT. Jou -
nal o S a is ical So wa e. 2015;p. accep ed o publica ion
17. Liaw A, Wiene M. Classi ica ion and Reg ession by andomFo es . R News. 2002; 2(3): 18–22. A ail-
able om: h p://CRAN.R-p ojec .o g/doc/Rnews/
18. Ve onese A, Lupini L, Consiglio J, Visone R, Fe acin M, Fo na i F, e al. Oncogenic ole o miR-483-3p
a he IGF2/483 locus. Cance Res. 2010; 70(8): 3140–3149. doi: 10.1158/0008-5472.CAN-09-4456
PMID: 20388800
19. Nakano K, Vousden KH. PUMA, a no el p oapop o ic gene, is induced by p53. Mol Cell. 2001; 7(3):
683–694. doi: 10.1016/S1097-2765(01)00214-3 PMID: 11463392
20. Dahiya R, Lee C, McCa ille J, Hu W, Kau G, Deng G. High equency o gene ic ins abili y o mic osa -
elli es in human p os a ic adenoca cinoma. In J Cance . 1997; 72(5): 762–767. doi: 10.1002/(SICI)
1097-0215(19970904)72:5%3C762::AID-IJC10%3E3.0.CO;2-B PMID: 9311591
21. Scel o RA, Schwienbache C, Ve onese A, G aman ie i L, Bolondi L, Que zoli P, e al. Loss o me hyla-
ion a ch omosome 11p15.5 is common in human adul umo s. Oncogene. 2002; 21(16): 2564–2572.
doi: 10.1038/sj.onc.1205336 PMID: 11971191
22. Wang Y, Zhang X, Li H, Yu J, Ren X. The ole o miRNA-29 amily in cance . Eu J Cell Biol. 2013; 92
(3): 123–128. doi: 10.1016/j.ejcb.2012.11.004 PMID: 23357522
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 16 / 17

23. Chen PS, Su JL, Cha ST, Ta n WY, Wang MY, Hsu HC, e al. miR-107 p omo es umo p og ession by
a ge ing he le -7 mic oRNA in mice and humans. J Clin In es . 2011; 121(9): 3442–3455. doi: 10.
1172/JCI45390 PMID: 21841313
24. Ba h D, Malho a R, Ra i B, Sindhu ani P. Mic oRNA le -7: an eme ging nex -gene a ion cance he a-
peu ic. Cu Oncol. 2010; 17(1): 70–80. doi: 10.3747/co. 17i1.356 PMID: 20179807
25. Cao Q, Yu J, Dhanaseka an SM, Kim JH, Mani RS, Tomlins SA, e al. Rep ession o E-cadhe in by he
polycomb g oup p o ein EZH2 in cance . Oncogene. 2008; 27(58): 7274–7284. doi: 10.1038/onc.2008.
333 PMID: 18806826
26. Gui J, Tian Y, Wen X, Zhang W, Zhang P, Gao J, e al. Se um mic oRNA cha ac e iza ion iden i ies
miR-885-5p as a po en ial ma ke o de ec ing li e pa hologies. Clin Sci. 2011; 120(5): 183–193. doi:
10.1042/CS20100297 PMID: 20815808
27. A anasye a EA, Mes dagh P, Kumps C, Vandesompele J, Ehemann V, Theissen J. Mic oRNA miR-
885-5p a ge s CDK2 and MCM5, ac i a es p53 and inhibi s p oli e a ion and su i al. Cell Dea h Di e .
2011; 18(6): 974–984. doi: 10.1038/cdd.2010.164 PMID: 21233845
28. Ei ing AM, Ha b JG, Ne iani P, Ga on C, Oaks JJ, Spizzo R, e al. miR-328 unc ions as an RNA decoy
o modula e hnRNP E2 egula ion o mRNA ansla ion in leukemic blas s. Cell. 2010; 140(5): 652–
665. doi: 10.1016/j.cell.2010.01.007 PMID: 20211135
29. Guo Z, Shao L, Zheng L, Du Q, Li P, John B, e al. miRNA-939 egula es human inducible ni ic oxide
syn hase pos ansc ip ional gene exp ession in human hepa ocy es. P oc Na l Acad Sci USA. 2012;
109(15): 5826–5831. doi: 10.1073/pnas.1118118109 PMID: 22451906
30. Loibl S, Buck A, S ank C, on Minckwi z G, Rolle M, Sinn HP, e al. The ole o ea ly exp ession o in-
ducible ni ic oxide syn hase in human b eas cance . Eu J Cance . 2005; 41(2): 265–271. doi: 10.
1016/j.ejca.2004.07.010 PMID: 15661552
31. Ekmekcioglu S, Elle ho s JA, P ie o VG, Johnson MM, B oemeling LD, G imm EA. Tumo iNOS p e-
dic s poo su i al o s age III melanoma pa ien s. In J Cance . 2006; 119(4): 861–866. doi: 10.1002/
ijc.21767 PMID: 16557582
32. Aal omaa SH, Lipponen PK, Vii anen J, Kankkunen JP, Ala-Opas MY, Kosma VM. The p ognos ic
alue o inducible ni ic oxide syn hase in local p os a e cance . BJU In . 2000; 234(239): 234–239. doi:
10.1046/j.1464-410x.2000.00787.x
33. Ho man AE, Liu R, Fu A, Zheng T, Slack F, Zhu Y. Ta ge ome P o iling, Pa hway Analysis and Gene ic
Associa ion S udy Implica e miR-202 in Lymphomagenesis. Cance epidemiol biom and p e . 2013;
22(3): 1–10.
34. Kuma MS, Lu J, Me ce KL, Golub TR, Jacks T. Impai ed mic oRNA p ocessing enhances cellula
ans o ma ion and umo igenesis. Na Gene . 2007; 39(5): 673–677. doi: 10.1038/ng2003 PMID:
17401365
35. Ve bee en J, Niemel¨a EH, Tu unen JJ, Will CL, Ra an i JJ, Lüh mann R, e al. An ancien mechanism
o splicing con ol: U11 snRNP as an ac i a o o al e na i e splicing. Mol Cell. 2010; 37(6): 821–833.
doi: 10.1016/j.molcel.2010.02.014 PMID: 20347424
36. Venables JP, Klinck R, Koh C, Ge ais-Bi d J, B ama d A, Inkel L, e al. Cance -associa ed egula ion
o al e na i e splicing. Na S uc u al & Mol Biol. 2009; 16: 670–676 doi: 10.1038/nsmb.1608
37. Misqui a-Ali CM, Cheng E, O’Hanlon D, Liu N, McGlade CJ, Tsao MS, e al. Global p o iling and molec-
ula cha ac e iza ion o al e na i e splicing e en s mis egula ed in lung cance . Mol and Cell Biol. 2010;
31(1): 138–150. doi: 10.1128/MCB.00709-10
38. Lapuk A, Ma H, Jakkula L, Ped o H, Bha acha ya S, Pu dom E, e al. Exon-le el mic oa ay analyses
iden i y al e na i e splicing p og ams in b eas cance . Mol Cance Res. 2010; 8(7): 961–974. doi: 10.
1158/1541-7786.MCR-09-0528 PMID: 20605923
39. Dixon AL, Liang L, Mo a MF, Chen W, Hea h S, Wong KC, e al. A genome-wide associa ion s udy o
global gene exp ession. Na Gene . 2007; 39(10): 1202–1207. doi: 10.1038/ng2109 PMID: 17873877
40. Gö ing HH, Cu an JE, Johnson MP, Dye TD, Cha leswo h J, Cole SA, e al. Disco e y o exp ession
QTLs using la ge-scale ansc ip ional p o iling in human lymphocy es. Na Gene . 2007; 39(10):
1208–1216. doi: 10.1038/ng2119 PMID: 17873875
MiRNA P o iles in Lymphoblas oid Cell Lines
PLOS ONE | DOI:10.1371/jou nal.pone.0127427 May 28, 2015 17 / 17