ORIGINAL RESEARCH
published: 15 Feb ua y 2019
doi: 10.3389/ onc.2019.00079
F on ie s in Oncology | www. on ie sin.o g 1Feb ua y 2019 | Volume 9 | A icle 79
Edi ed by:
Adam Finn Binde ,
Thomas Je e son Uni e si y,
Uni ed S a es
Re iewed by:
Michael Diaman idis,
Uni e si y Hospi al o La issa, G eece
Je y Polesel,
Cen o di Ri e imen o Oncologico di
A iano (IRCCS), I aly
*Co espondence:
Ad ián Mosque a O guei a
[email p o ec ed]
Special y sec ion:
This a icle was submi ed o
Hema ologic Malignancies,
a sec ion o he jou nal
F on ie s in Oncology
Recei ed: 23 No embe 2018
Accep ed: 29 Janua y 2019
Published: 15 Feb ua y 2019
Ci a ion:
Mosque a O guei a A, An elo
Rod íguez B, Alonso Vence N,
Bendaña López Á, Díaz A ias JÁ, Díaz
Va ela N, González Pé ez MS, Pé ez
Encinas MM and Bello López JL
(2019) Time o T ea men P edic ion in
Ch onic Lymphocy ic Leukemia Based
on New T ansc ip ional Pa e ns.
F on . Oncol. 9:79.
doi: 10.3389/ onc.2019.00079
Time o T ea men P edic ion in
Ch onic Lymphocy ic Leukemia
Based on New T ansc ip ional
Pa e ns
Ad ián Mosque a O guei a1,2,3*, Bea iz An elo Rod íguez1,2,3, Na alia Alonso Vence1,2,
Ángeles Bendaña López1,2, José Ángel Díaz A ias1,2, Nicolás Díaz Va ela2,
Ma a Sonia González Pé ez1,2, Manuel Ma eo Pé ez Encinas1,3 and
José Luis Bello López1,2,3
1Heal h Resea ch Ins i u e o San iago de Compos ela (IDIS), San iago de Compos ela, Spain, 2Di ision o Hema ology,
Complexo Hospi ala io Uni e si a io de San iago de Compos ela, SERGAS, San iago de Compos ela, Spain, 3Depa men o
Medicine, Uni e si y o San iago de Compos ela, San iago de Compos ela, Spain
Ch onic lymphocy ic leukemia (CLL) is he mos equen lymphop oli e a i e synd ome in
wes e n coun ies. CLL e olu ion is equen ly indolen , and ea men is mos ly ese ed
o hose pa ien s wi h signs o symp oms o disease p og ession. In his wo k, we used
RNA sequencing da a om he In e na ional Cance Genome Conso ium CLL coho
o de e mine new gene exp ession pa e ns ha co ela e wi h clinical e olu ion.We
de e mined ha a 290-gene exp ession signa u e, in addi ion o immunoglobulin hea y
chain a iable egion (IGHV) mu a ion s a us, s a i ies pa ien s in o ou g oups wi h
no ably di e en ime o i s ea men . This inding was con i med in an independen
coho . Simila ly, we p esen a machine lea ning algo i hm ha p edic s he need o
ea men wi hin he i s 5 yea s ollowing diagnosis using exp ession da a om 2,198
genes. This p edic o achie ed 90% p ecision and 89% accu acy when classi ying
independen CLL cases. Ou indings indica e ha CLL p og ession isk la gely co ela es
wi h pa icula ansc ip omic pa e ns and pa es he way o he iden i ica ion o high- isk
pa ien s who migh bene i om p omp he apy ollowing diagnosis.
Keywo ds: ch onic lymphocy ic leukemia, ime o ea men p edic ion, gene exp ession, RNAseq, machine
lea ning, p ognos ic ac o s, IGHV
INTRODUCTION
Ch onic lymphocy ic leukemia (CLL) is a low-g ade B-cell lymphop oli e a i e disease wi h an
es ima ed yea ly incidence in wes e n coun ies o abou 6.9 cases pe 100,000 people (1) and
ema kable a ia ion be ween aces. The incidence o CLL is highe in men han in women and i
inc eases p og essi ely om he age o 35 un il he las decades o li e (2). Cu en ly, CLL ea men
is delayed un il disease p og ession (bone ma ow ailu e, o ganomegaly, gene al symp oms, o
high-g ade lymphoma ans o ma ion) and in he case o e ac o y au oimmune phenomena
(3,4). Ne e heless, wi h he ad en o new a ge ed ea men s such as ib u inib (5), idelalisib
(6), and ene oclax (7), i is emp ing o specula e ha some indi iduals could bene i om ea ly
in e en ion immedia ely ollowing diagnosis, when he umo al mass is smalle and pa ien s ha e
a be e physical condi ion. Thus, imp o ed isk s a i ica ion o pa ien s wi h CLL is needed.
Mosque a O guei a e al. T ansc ip omics o Time o The apy in CLL
Recen ad ances in CLL genomics ha e disco e ed new
d i e s o disease, many o which a e associa ed wi h a di e en
clinical e olu ion. Dele ions (6p21, 6q15, 11q, 14q24, 15q15,
17p, 18p, and 20p; gains in 2p16, 5q24, and 8q24), isomy
12, and gene mu a ions (TP53,ATM,NOTCH1,SF3B1,BIRC3,
BRAF,POT1,ZNF292,NFKB2,MGA,IRF4,DDX3X,ZMYM3,
and FUBP1)ha e been epea edly obse ed in he CLL genome
and a e linked o apid disease p og ession (8) Ne e heless,
immunoglobulin hea y chain a iable egion (IGHV) mu a ion
s a us, which is an indi ec measu e o he umo lymphocy es’
ma u a ion s age (9), is among he mos impo an single
p edic i e ac o known o da e (10). IGHV unmu a ed pa ien s
show ema kably wo se p ognosis han IGHV mu a ed pa ien s
(10,11) and only a ew o he genomic ac o s ha e p o en o
be associa ed wi h clinical e olu ion independen o his a iable.
Lymphocy e ma u a ion is such an impo an indica o ha DNA
me hyla ion s a us has been used o classi y CLL in o h ee
di e en g oups ha esemble di e en B cell ma u a ion s ages
(nai e B cell, in e media e, and memo y B cell). This classi ica ion
was shown o ou pe o m IGHV s a us a p edic ing ime o i s
ea men (TTT) (12).
Mu a ions, genomic abe a ions, and DNA me hyla ion
pa e ns induce ansc ip omic changes ha can be measu ed
using RNA sequencing (RNAseq), a echnique ha o e s an
oppo uni y o iden i y new bioma ke s o disease p og ession
and d ug esponse p edic ion (13–15). In ac , p e ious e o s
o imp o e CLL isk s a i ica ion based on RNAseq da a ha e
demons a ed imp essi e esul s (16), bu he clinical applica ion
is di icul due o he expense o ex ensi e echnical and
bioin o ma ics e o s. The e o e, he e is a need o smalle
ansc ip omics pa e ns co ela ed wi h disease e olu ion o
medical use.
In his s udy, we pe o med machine-lea ning based Gaussian
mix u e model clus e ing on a subg oup o genes signi ican ly
associa ed wi h TTT in o de o iden i y ansc ip ional clus e s
wi h clinical implica ions. We s udied TTT due o he lack
o ea men uni o mi y in he In e na ional Cance Genome
Conso ium (ICGC) CLL coho and because i is a a iable
associa ed wi h o e all su i al (17). We es ed ou esul s on
a 196 pa ien coho and alida ed i s clinical signi icance in an
independen 79 pa ien coho . The o e all esul s delinea ed wo
IGHV-independen ansc ip ional clus e s ha s a i y pa ien s
acco ding o hei isk o ea men ini ia ion. Fu he mo e,
we demons a ed ha machine lea ning algo i hms using gene
exp ession da a can p edic pa ien need o ea men in
he i s 5 yea s ollowing diagnosis. We an icipa e ha ou
indings will imp o e he iden i ica ion o high- isk CLL pa ien s
ollowing diagnosis.
MATERIALS AND METHODS
Da a Sou ces and Pa ien Cha ac e is ics
We applied o access o he ICGC’s CLL sequencing da a
(18) deposi ed in he Eu opean Genome-Phenome Da abase
(EGA) (19). The Da a Access Commi ee app o ed access o
his da a unde DACO-1040945. Two CLL RNA-seq coho s
TABLE 1 | Pa ien cha ac e is ics o he es and alida ion coho s.
Ca ego y Tes coho Valida ion coho
Cases 196 79
Age a diagnosis (median) 63 62
Sex (% males) 60.70% 69.62
MBL 11.20% 3.79%
Bine A 77.44% 91.13%
Bine B 7.18% 3.79%
Bine C 4.10% 1.26%
IGHV unmu a ed 32.65% 43%
SLL 2.55% 2.04%
P opo ion o p og essions in he i s
5 yea s since diagnosis
31.12% 31.64%
we e uploaded in wo s ages wi h he ollowing accession codes:
EGAD00001001443 and EGAD00001000258.
The i s coho (EGAD00001001443, he ea e s udy coho )
con ains RNAseq da a and om CLL-pu i ied cells o 196
indi iduals along wi h clinical da a. The coho was composed
o 169 CLL, 22 monoclonal B cell lymphocy osis (MBL), and
i e small lymphocy ic lymphoma (SLL) samples. The e we e 132
IGHV mu a ed cases and 64 IGHV unmu a ed cases in 119 males
and 77 emales. By s aging a diagnosis, he e we e 22 MBL cases,
151 Bine S age A cases, 14 Bine S age B cases, and 8 Bine C
s age cases.
The second coho (EGAD00001000258, he ea e alida ion
coho ) is composed o RNAseq da a o CLL-pu i ied cells om
98 indi iduals, o which 79 (55 males and 24 emales) ha e
publicly a ailable pheno ypic in o ma ion. In his coho he e
we e 72 CLL, 4 SLL, and 3 MBL samples. 45 o he pa ien s
had mu a ed IGHV and 34 had unmu a ed IGHV. By s aging a
diagnosis, he e we e 3 MBL, 72 Bine S age A, 3 Bine S age B,
and 1 Bine S age C cases.
A summa y o he pa ien cha ac e is ics o bo h coho s can
be consul ed in Table 1.
Da a P ep ocessing and Alignmen
RNAseq pai ed-end da a we e ob ained om Illumina pai ed-
end sequencing pe o med by he ICGC CLL conso ium
as desc ibed by Fe ei a e al. (16) Illumina adap e s we e
emo ed using cu adap (20) and alignmen o he human
e e ence genome (GRCh37) was pe o med using Hisa 2
(21) wi h de aul speci ica ions. We used he Hisa 2-p o ided
Hie a chical G aph FM index o GRCh37 wi h SNP and
Ensembl ansc ip in o ma ion. Bam iles we e so ed and
indexed using sam ools (22).
Gene Exp ession Es ima ion
RNAseq bam iles we e p ocessed in R(23) acco ding o he
RNAseq gene exp ession p o ocol de eloped by Lo e e al. (24)
B ie ly, bam iles we e ead using Rsam ools, (25) ollowed by
gene-le el exp ession es ima ion using he Summa izeO e laps
unc ion om he GenomicAlignmen s package. (26) Gene
models in GTF o ma we e downloaded om Ensembl
F on ie s in Oncology | www. on ie sin.o g 2Feb ua y 2019 | Volume 9 | A icle 79
Mosque a O guei a e al. T ansc ip omics o Time o The apy in CLL
FIGURE 1 | Hea map showing he ank- ans o med dis ibu ion o exp ession alues o he 290 genes in he s udy coho . Red-labeled samples on he le ba
pe ain o C1 and blue-labeled samples pe ain o C2.
(GRCh37.75 e sion) (27). Genes wi h a median ead coun
below one we e disca ded.
S a is ical Analysis
We analyzed gene exp ession associa ion wi h CLL’s TTT using
cox eg ession implemen ed in he su i al package (28,29). In
his model we included he co a ia es dono sex and CLL s age
(MBL, Bine S age A, Bine S age B, and Bine S age C). Time o
T ea men was calcula ed as he pe iod be ween CLL diagnosis
and he ini ia ion o he i s ea men o CLL. The day o las
ollow-up was used o igh censo ing he da a o pa ien s wi h
incomple e ollow-up.
Clus e ing was pe o med using he Mclus package (30) wi h
de aul pa ame e s. B ie ly, Mclus in e s he likelies da a clus e s
based on Gaussian Mix u e Modeling i ed by an Expec a ion-
Maximiza ion (GMM-EM) algo i hm.
Those genes wi h signi ican associa ion wi h TTT in he
s udy coho (cox eg ession alse disco e y a e [FDR] below
5%) we e selec ed as ou ini ial lis o genes. Va iable selec ion
was pe o med by adding one new gene in p- alue ascending
o de o he model (s a ing wi h he i s wo mos signi ican
genes un il eaching he op a 2,198 genes [FDR<5%]) and
compu ing he mos likely clus e s. Fo he sake o simplici y, we
disca ded he 25% leas a iable genes, he 50% leas exp essed
genes and hose wi h a high (>0.9) Spea man’s ank co ela ion
wi h any o he gene in he inpu da a. In he case o a highly
co ela ed pai o genes, he one wi h he lowes p- alue was
disca ded. In each i e a ion we o ced Mclus o calcula e he wo
mos likely g oups o samples in ou da a, and o selec he bes
model acco ding o he maximal Bayesian In o ma ion C i e ion
(BIC). Associa ion wi h TTT calcula ed using cox eg ession
(su i al package), including IGHV mu a ion s a us as co a ia e
F on ie s in Oncology | www. on ie sin.o g 3Feb ua y 2019 | Volume 9 | A icle 79
Mosque a O guei a e al. T ansc ip omics o Time o The apy in CLL
FIGURE 2 | Kaplan-Meie su i al plo s. The uppe plo s show he associa ion o C1 ( ed cu e) and C2 (blue cu e) wi h TTT in he s udy (le ) and alida ion coho s
( igh ). Co esponding p- alues a e 1.7 ×10−6and 1.3 ×10−4. The lowe plo s show he associa ion wi h TTT s a i ied by IGHV mu a ion s a us in he s udy (le )
and alida ion coho s ( igh ). The blue line indica es C2 samples wi h mu a ed IGHV, he pu ple line indica es C2 samples wi h unmu a ed IGHV, he ed line
indica es C1 samples wi h mu a ed IGHV, and he g een line e e s o C1 samples wi h unmu a ed IGHV.
in each i e a ion. P- alue adjus men was pe o med wi h he
Bon e oni me hod.
Machine Lea ning Ensembl Cons uc ion
Fo IGHV s a us and need o ea men a 5 yea s p edic ion
we an boos ed ees analysis using BigML applica ions (31)
wi h a 2,000 ee node h eshold. We chose 5 yea s due o he
ollowing easons: (1) i is impo an o di e which pa ien s will
ha e p og ession in he i s yea s since diagnosis; and (2) he
numbe o cases p og essing in ea lie yea s was oo small in
o de o ain a good classi ica o . Va ying pe cen ages o lea ning
a es we e es ed. The bes model was selec ed based on ecei e
ope a ing cha ac e is ic (ROC) cu es, P ecision-Recall cu es,
and Kolmogo o -Smi no s a is ics.
RESULTS
Genes Associa ed Wi h Time o T ea men
and Clus e iza ion
A cox eg ession model was cons uc ed wi h gene exp ession,
dono sex and CLL s age a diagnosis as independen a iables.
2,198 genes we e ound o be signi ican ly associa ed wi h TTT
(FDR <5%) in he s udy coho .
Pa ien clus e iza ion based on gene exp ession da a
using a GMM-EM algo i hm e ie ed 19 se s o genes
F on ie s in Oncology | www. on ie sin.o g 4Feb ua y 2019 | Volume 9 | A icle 79
Mosque a O guei a e al. T ansc ip omics o Time o The apy in CLL
FIGURE 3 | Hea map showing he ank- ans o med dis ibu ion o exp ession alues o he 290 genes in he alida ion coho . Red-labeled samples on he le ba
pe ain o C1 and blue-labeled samples pe ain o C2.
ha clus e ed samples in o wo g oups wi h signi ican
associa ions wi h TTT when adjus ed o IGHV s a us
(Bon e oni-adjus ed p- alue <0.01) (Supplemen al Table 1).
The mos signi ican clus e (clus e 2) con ained 290
ansc ip s (Figu e 1,Supplemen al Table 2) and achie ed
an associa ion p- alue o 6.4 ×10−7(Bon e oni p- alue
1.4 ×10−3) wi h he TTT a iable adjus ed o IGHV
mu a ion s a us (Figu e 2). A signi ican associa ion was
con i med in he alida ion coho (IGHV adjus ed p- alue 3.05
×10−3) (Figu es 2,3).
Acco ding o he selec ed classi ica o , pa ien s in clus e
wo (C2) had a mo e a o able p ognosis han pa ien s in
clus e one (C1) (Haza d Ra ios (HR) o −1.70 and−1.41 in
he es and alida ion coho s, espec i ely), independen ly
o IGHV mu a ion s a us. Among he s udy coho , oughly
36.7% o pa ien s belonged o C2, while 34.1% o pa ien s
in he alida ion coho clus e ed wi hin C2. C2 in ol ed
51.5% o IGHV-mu a ed pa ien s and 6.4% o IGHV-unmu a ed
pa ien s in he s udy coho , as well as 55.5% o IGHV-
mu a ed pa ien s and 5.8% o IGHV-unmu a ed pa ien s in he
alida ion coho .
Machine Lea ning o T ea men F ee
Su i al P edic ion
We we e in e es ed in a machine lea ning (ML) classi ie ha
could p edic which pa ien s would equi e CLL he apy in he
i s yea s ollowing diagnosis. We cons uc ed model ensembles
wi h all genes associa ed wi h TTT in Cox eg ession a a
FDR o 5%. We also es ed di e en lea ning a es (0.5, 1,
2.5, 5, and 10%). 222 pa ien s had a ollow-up pe iod >5yea s
o had been ea ed in he i s 5 yea s ollowing diagnosis,
and we di ided hem in o a aining se (80% o pa ien s,
F on ie s in Oncology | www. on ie sin.o g 5Feb ua y 2019 | Volume 9 | A icle 79
Mosque a O guei a e al. T ansc ip omics o Time o The apy in CLL
FIGURE 4 | ROC cu e o he boos ed- ee Ensembl model o 5 yea ea men need p edic ion (uppe le ). Kolmogo o -Smi no plo o he same model (uppe
igh ). P ecision-Recall plo o he 5 yea no - ea ed (lowe le ) and ea ed (lowe igh ) pa ien s acco ding o he same model. Whi e do s in each g aph indica e
he p obabili y h eshold (in his case 50%), which is he poin e lec ing he bes classi ica ion accu acy o he pa ien s.
composed o 146 pa ien s om he s udy coho and 31 pa ien s
om he alida ion coho ) and a es se (20% o pa ien s,
composed o 45 pa ien s om he alida ion coho ). ROC AUC
and P ecision-Recall AUC plo s we e e alua ed o selec he
bes esul s.
The bes model used a 2.5% lea ning a e and 2,000 ee
nodes. I achie ed 90% p ecision a iden i ying pa ien s ha
needed ea men in 5 yea s wi h 69.23% ecall, and 88.57%
p ecision a iden i ying hose pa ien s ha did no equi e
ea men in 5 yea s wi h 96.88% ecall. We only de ec ed
1 alse posi i e case (3.1% False Posi i e Ra e) and 4 alse
nega i es (30% False Nega i e Ra e). A e age p ecision was
89.29%, accu acy was 88.89% and ROC a ea unde he cu e
(AUC) was 0.911 (Figu e 4 and Table 2). P ecision-Recall AUC
was 0.860 and 0.959 o p edic ing which pa ien s would o would
no need ea men wi hin his pe iod, espec i ely. In each case,
he esul s mos ly o e lapped wi h he a ea unde he con ex
hull (AUCH).
DISCUSSION
The main aim o his s udy was o iden i y new ansc ip omic
pa e ns in o de o imp o e CLL pa ien isk s a i ica ion.
We used he GMM-EM algo i hm o s a i y pa ien s in wo
clus e s wi h ema kably di e en clinical beha io based on
he exp ession o 290 genes, and we obse ed ha his pa e n
was independen o IGHV mu a ion s a us. In e es ingly, we
iden i ied a g oup o CLL pa ien s wi h mu a ed IGHV and
F on ie s in Oncology | www. on ie sin.o g 6Feb ua y 2019 | Volume 9 | A icle 79
Mosque a O guei a e al. T ansc ip omics o Time o The apy in CLL
TABLE 2 | Con usion Ma ix o he boos ed- ee Ensembl model p edic ing he
5 yea need o ea men .
5 yea ea .
need
No needs
ea .
Needs
ea .
Ac ual Recall(%)
No needs ea . 31 1 32 96.88
needs ea . 4 9 13 69.23
p edic ed 35 10 45 83.05**
p ecision 88.57% 90.00% 89.29%* 88.89***
*A e age p ecision.
**A e age ecall.
***Accu acy p obabili y h eshold =50%.
a low- isk ansc ip omic p o ile ha only need ea men in
app oxima ely 25% o he cases du ing disease e olu ion. Two
addi ional g oups (one composed o pa ien s wi h mu a ed IGHV
and a high- isk ansc ip omic p o ile and he second composed
o unmu a ed IGHV pa ien s wi h a low- isk ansc ip omic
p o ile) ha e simila in e media e e olu ion, while a inal g oup
(composed o pa ien s wi h unmu a ed IGHV and an ad e se
ansc ip omic p o ile) has he highes p obabili y o ea men
need in he i s yea s ollowing diagnosis. These esul s a e
conco dan wi h p e ious epo s in he ield. Fo example,
Yepes e al. (32) epo ed a di ision o CLL cases in wo
g oups based on mic oa ay ansc ip ome cha ac e iza ion
h ough unsupe ised clus e ing analysis, which was alida ed
in 4 independen coho s. Simila ly, F iedman e al. (33)
desc ibed a 180 p obe classi ie based on mic oa ay da a
ha also di ided wo clus e s o CLL pa ien s independen ly
o IGHV mu a ion s a us. Ou indings a e also simila o
hose published by Fe ei a e al. (16), who desc ibed wo
gene exp ession clus e s ha show IGHV mu a ion-independen
associa ion wi h TTT using an ea ly elease o he ICGC
CLL coho . Ne e heless, he e a e ema kable di e ences
be ween ou analysis and ha o Fe ei a e al, Yepes e al.
and F iedman e al. Fi s ly, ou clus e iza ion is based on
a ansc ip ional pa e n o a small subg oup o genes ha
acili a es i s u u e applicabili y, whils hose o Fe ei a
e al. and Yepes e al. a e based on whole ansc ip ome
analysis. Secondly, ou classi ie is based on RNAseq da a,
a echnology ha has ou pe o med mic oa ay analysis in
mos ields. Wi h he use RNAseq i will be possible o
couple ansc ip ome clus e iza ion wi h a ge ed gene mu a ion
de ec ion, s e eo yped B cell ecep o exp ession o IGHV
hype mu a ion s a us analysis.
We also desc ibe a no el a i icial in elligence algo i hm
ha can p edic a CLL pa ien ’s need o he apy du ing
he i s 5 yea s ollowing diagnosis wi h high p ecision and
accu acy. This is in line wi h o he ML applica ions o oncologic
malignancies ha a e s a ing o change pa adigms in pa ien isk
s a i ica ion and d ug esponse p edic ion. Fo example, Aziz
e al. (34) ecen ly epo ed he iden i ica ion o a ML model
ha in eg a es clinical and genomic da a om pa ien s wi h
myelodysplas ic synd ome (MDS). This model ou pe o med
all commonly used p edic ion models in he ield o MDS.
Simila ly, Youse i e al. (35) used bayesian-op imized deep
lea ning o su i al p edic ion in pan-cance analysis, showing
no only be e pe o mance han o he s a e-o - he-a me hods,
bu also imp o ed p edic abili y o cance su i al h ough
ans e lea ning in di e en ypes o cance genomic da a.
Thus, i is likely ha ML-d i en algo i hms applied o genomic
and ansc ip omic da a will be used in he nea u u e o
he iden i ica ion o “smolde ing” CLLs ha may bene i om
ea ly in e en ion.
RNAseq is a powe ul echnique ha can sequence he
whole ansc ip ome a an inc easingly lowe cos . Ta ge ed
RNAseq is being de eloped o clinical applica ion, wi h
he addi ional possibili y o es ing o gene mu a ions and
usion genes in he same echnique. The e o e, de ining
ep oducible gene exp ession pa e ns wi h clinical implica ions
is a s a egy ha can close he gap be ween esea ch and he
clinical p ac ice. He e we p esen pa e ns o gene exp ession
ha can imp o e CLL pa ien isk s a i ica ion wi h a
ela i ely small se o he ansc ip ome. These esul s may
pa e he way o he design o new ea men s a egies
in ol ing ea ly CLL ea men in high- isk pa ien s be o e
disease p og ession.
AUTHOR CONTRIBUTIONS
AMO designed he s udy pe o med esea ch. AMO, BAR, JDA,
and JBL analyzed he da a. AMO w o e he pape . NAV, ABL,
NDV, MGP, and MPE e iewed he pape .
FUNDING
The publica ion cos s associa ed wi h his manusc ip ha e been
paid by Roche Pha maceu icals. The unde played no ole in he
s udy design, he collec ion, analysis o in e p e a ion o da a, he
w i ing o his pape o he decision o submi i o publica ion.
ACKNOWLEDGMENTS
The au ho s hank he ICGC conso ium o sha ing he genomic
da a and he Cen o de Supe compu ación de Galicia (CESGA)
o app o ing he access o in o ma ics acili ies wi h echnical
suppo . We would also like o hank BigML Inc., pa icula ly
F ancisco J. Ma ín, o access o machine lea ning applica ions.
The con en o his pape is pa o he doc o al hesis o AMO
o ob ain a PhD a he Depa men o Medicine, Uni e si y o
San iago de Compos ela.
SUPPLEMENTARY MATERIAL
The Supplemen a y Ma e ial o his a icle can be ound
online a : h ps://www. on ie sin.o g/a icles/10.3389/ onc.
2019.00079/ ull#supplemen a y-ma e ial
Supplemen al Table 1 | This able ep esen s he o iginal 19 clus e s wi h
Bon e oni p- alues <0.01.
Supplemen al Table 2 | Lis o he 290 genes used o C1 s. C2 clus e iza ion.
F on ie s in Oncology | www. on ie sin.o g 7Feb ua y 2019 | Volume 9 | A icle 79
Mosque a O guei a e al. T ansc ip omics o Time o The apy in CLL
REFERENCES
1. Zhao Y, Wang Y, Ma S. Racial di e ences in ou leukemia sub ypes:
comp ehensi e desc ip i e epidemiology. Sci Rep. (2018) 8:548.
doi: 10.1038/s41598-017-19081-4
2. Do es GM, Ande son WF, Cu is RE, Landg en O, Os oumo a
E, Bluhm EC, e al. Ch onic lymphocy ic leukaemia and small
lymphocy ic lymphoma: o e iew o he desc ip i e epidemiology.
B J Haema ol. (2007) 139:809–19. doi: 10.1111/j.1365-2141.2007.0
6856.x
3. Mozas P, Ri as-Delgado A, Baumann T, Villamo N, O iz-Maldonado V,
Ayme ich M, e al. Analysis o c i e ia o ea men ini ia ion in pa ien s
wi h p og essi e ch onic lymphocy ic leukemia. Blood Cance J. (2018) 8:10.
doi: 10.1038/s41408-017-0044-5
4. Eichho s B, Robak T, Mon se a E, Ghia P, Hillmen P, Hallek M, e al.
Ch onic lymphocy ic leukaemia: ESMO Clinical P ac ice Guidelines o
diagnosis, ea men and ollow-up. Ann Oncol. (2015) 26 (Suppl. 5): 78–84.
doi: 10.1093/annonc/md 303
5. Bu ge JA, Tedeschi A, Ba PM, Robak T, Owen C, Ghia P, e al. Ib u inib as
Ini ial he apy o pa ien s wi h ch onic lymphocy ic leukemia. N Engl J Med.
(2015) 373:2425–37. doi: 10.1056/NEJMoa1509388
6. B own JR, By d JC, Cou e SE, Benson DM, Flinn IW, Wagne -Johns on
ND, e al. Idelalisib, an inhibi o o phospha idylinosi ol 3-kinase p110δ, o
elapsed/ e ac o y ch onic lymphocy ic leukemia. Blood (2014) 123:3390–7.
doi: 10.1182/blood-2013-11-535047
7. Robe s AW, Da ids MS, Pagel JM, Kahl BS, Pu ada SD, Ge eci ano JF, e al.
Robe s AW, Da ids MS, Pagel JM, e al. Ta ge ing BCL2 wi h ene oclax
in elapsed ch onic lymphocy ic leukemia. N Engl J Med. (2016) 374:311–22.
doi: 10.1056/NEJMoa1513257
8. Puen e XS, Beà S, Valdés-Mas R, Villamo N, Gu ié ez-Ab il J, Ma ín-Sube o
JI, e al. Non-coding ecu en mu a ions in ch onic lymphocy ic leukaemia.
Na u e (2015) 526:519–24. doi: 10.1038/na u e14666
9. Oakes CC, Sei e M, Asseno Y, Gu L, P zekopowi z M, Ruppe AS, e al.
DNA me hyla ion dynamics du ing B cell ma u a ion unde lie a con inuum
o disease pheno ypes in ch onic lymphocy ic leukemia. Na Gene . (2016)
48:253–64. doi: 10.1038/ng.3488
10. Damle RN, Wasil T, Fais F, Ghio o F, Vale o A, Allen SL, e al. Ig V
gene mu a ion s a us and CD38 exp ession as no el p ognos ic indica o s in
ch onic lymphocy ic leukemia. Blood (1999) 94:1840–7.
11. Hamblin TJ, Da is Z, Ga dine A, Oscie DG, S e enson FK. Unmu a ed Ig
V(H) genes a e associa ed wi h a mo e agg essi e o m o ch onic lymphocy ic
leukemia. Blood (1999) 94:1848–54.
12. Quei ós AC, Villamo N, Clo G, Ma inez-T illos A, Kulis M, Na a o A,
e al. A B-cell epigene ic signa u e de ines h ee biologic subg oups o ch onic
lymphocy ic leukemia wi h clinical impac . Leukemia (2015) 29:598–605.
doi: 10.1038/leu.2014.252
13. Maag JLV, Fishe OM, Le e -Mignon A, Kaczo owski DC, Thomas ML,
Hussey DJ, e al. No el abe a ions unco e ed in ba e ’s esophagus and
esophageal adenoca cinoma using whole ansc ip ome sequencing. Mol
Cance Res. (2017) 15:1558–69. doi: 10.1158/1541-7786.MCR-17-0332
14. Wang Q, Gan H, Chen C, Sun Y, Chen J, Xu M, e al. Iden i ica ion
and alida ion o a 44-gene exp ession signa u e o he classi ica ion
o enal cell ca cinomas. J Exp Clin Cance Res. (2017) 36:176.
doi: 10.1186/s13046-017-0651-9
15. Zhang YH, Huang T, Chen L, Xu Y, Hu Y, Hu LD, e al. Iden i ying and
analyzing di e en cance sub ypes using RNA-seq da a o blood pla ele s.
Onco a ge (2017) 8:87494–511. doi: 10.18632/onco a ge .20903
16. Fe ei a PG, Ja es P, Rico D, Gómez-López G, Ma ínez-T illos A, Villamo N,
e al. T ansc ip ome cha ac e iza ion by RNA sequencing iden i ies a majo
molecula and clinical subdi ision in ch onic lymphocy ic leukemia. Genome
Res. (2014) 24:212–26. doi: 10.1101/g .152132.112
17. Ammann EM, Shana el TD, W igh KB, McDowell BD, Link BK,
Ch ischilles EA. Upda ing su i al es ima es in pa ien s wi h ch onic
lymphocy ic leukemia o small lymphocy ic lymphoma (CLL/SLL) based
on ea men - ee in e al leng h. Leuk Lymphoma (2018) 59:643–9.
doi: 10.1080/10428194.2017.1349905
18. Ramsay AJ, Ma ínez-T illos A, Ja es P, Rod íguez D, Kwa ciak A,
Quesada V. Nex -gene a ion sequencing e eals he sec e s o he
ch onic lymphocy ic leukemia genome. Clin T ansl Oncol. (2013) 15:3–8.
doi: 10.1007/s12094-012-0922-z
19. Lappalainen I, Almeida-King J, Kumandu i V, Sen A, Spalding JD, U -
Rehman S, e al. The Eu opean Genome-phenome A chi e o human
da a consen ed o biomedical esea ch. Na Gene . (2015) 47:692–5.
doi: 10.1038/ng.3312
20. Ma in M. Cu adap Remo es Adap e Sequences F om High-Th oughpu
Sequencing Reads. EMBne .jou nal, [S.l.], . 17, n. 1, p. pp. 10-12, may. (2011)
ISSN 2226-6089. A ailable online a : <h p://jou nal.embne .o g/index.php/
embne jou nal/a icle/ iew/200>(accessed Feb ua y 15, 2018).
21. Kim D, Langmead B, Salzbe g SL. HISAT: a as spliced aligne
wi h low memo y equi emen s. Na Me hods. (2015) 12:357–60.
doi: 10.1038/nme h.3317
22. Li H, Handsake B, Wysoke A, Fennell T, Ruan J, Home N, e al.
The Sequence Alignmen /Map o ma and SAM ools. Bioin o ma ics (2009)
25:2078–9. doi: 10.1093/bioin o ma ics/b p352
23. R De elopmen Co e Team R: A Language and En i onmen o S a is ical
Compu ing. Vienna, Aus ia: he R Founda ion o S a is ical Compu ing.
ISBN: 3-900051-07-0. A ailable online a : h p://www.R-p ojec .o g/ (2011)
24. Lo e MI, Ande s S, Kim V, Hube W. RNA-seq Wo k low: Gene-le el
Explo a o y Analysis and Di e en ial Exp ession. Bioconduc o . A ailable
online a : h ps://www.bioconduc o .o g/help/wo k lows/ naseqGene/ (2017)
25. Mo gan M, Pagès H, Obenchain V, Hayden N. Rsam ools: Bina y alignmen
(BAM), FASTA, Va ian Call (BCF), and Tabix File Impo . R package
e sion 1.30.0, A ailable online a : h p://bioconduc o .o g/packages/ elease/
bioc/h ml/Rsam ools.h ml (2017)
26. Law ence M, Hube W, Pagès H, Aboyoun P, Ca lson M, Gen leman R. e al.
So wa e o compu ing and anno a ing genomic anges. PLoS Compu Biol.
(2013) 9:e1003118. doi: 10.1371/jou nal.pcbi.1003118
27. Ya es A, Akanni W, Amode MR, Ba ell D, Billis K, Ca alho-Sil a
D, e al. Ensembl 2016. Nucleic Acids Res. (2016) 44(D1):D710–6.
doi: 10.1093/na /gk 1157
28. The neau T (2015). A Package o Su i al Analysis in S. e sion 2.38,
A ailable online a : h ps://CRAN.R-p ojec .o g/package=su i al
29. The neau TM, G ambsch PM. Modeling Su i al Da a: Ex ending he
Cox Model. New Yo k, NY: Sp inge (2000) doi: 10.1007/978-1-475
7-3294-8
30. Sc ucca L, Fop M, Mu phy TB, Ra e y AE. mclus 5: clus e ing, classi ica ion
and densi y es ima ion using gaussian ini e mix u e models. R J. (2016)
8:289–317.
31. BigML is Machine Lea ning made easy, iewed 15 Feb ua y 2018, A ailable
online a : h ps://bigml.com
32. Yepes S, To es MM, And ade RE. Clus e ing o exp ession da a in ch onic
lymphocy ic leukemia e eals new molecula subdi isions. PLoS ONE (2015)
10:e0137132. doi: 10.1371/jou nal.pone.0137132
33. F iedman DR, Weinbe g JB, Ba y WT, Goodman BK, Volkheime AD, Bond
KM, e al. A genomic app oach o imp o e p ognosis and p edic he apeu ic
esponse in ch onic lymphocy ic leukemia. Clin Cance Res. (2009) 15:6947–
55. doi: 10.1158/1078-0432.CCR-09-1132
34. Nazha A, Kom okji RS, Ba na d J, Al-Issa, K., Pad on, E., Madana , Y. F, e al.
A pe sonalized p edic ion model o isk s a i y pa ien s wi h myelodysplas ic
synd omes (MDS). Blood (2017) 130(Suppl 1):160.
35. Youse i S, Am ollahi F, Amgad M, Dong C, Lewis JE, Song C, e al. P edic ing
clinical ou comes om la ge scale cance genomic p o iles wi h deep su i al
models. Sci Rep. (2017) 7:11707. doi: 10.1038/s41598-017-11817-6
Con lic o In e es S a emen : The au ho s decla e ha he esea ch was
conduc ed in he absence o any comme cial o inancial ela ionships ha could
be cons ued as a po en ial con lic o in e es .
Copy igh © 2019 Mosque a O guei a, An elo Rod íguez, Alonso Vence, Bendaña
López, Díaz A ias, Díaz Va ela, González Pé ez, Pé ez Encinas and Bello López.
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 (CC BY). The use, dis ibu ion o ep oduc ion in o he o ums
is pe mi ed, p o ided he o iginal au ho (s) and he copy igh owne (s) a e c edi ed
and ha he o iginal publica ion in his jou nal is ci ed, in acco dance wi h accep ed
academic p ac ice. No use, dis ibu ion o ep oduc ion is pe mi ed which does no
comply wi h hese e ms.
F on ie s in Oncology | www. on ie sin.o g 8Feb ua y 2019 | Volume 9 | A icle 79
