Deconvolution of bulk blood eQTL effects into immune cell subpopulations.

METHODOLOGY ARTICLE Open Access Decon olu ion o bulk blood eQTL e ec s in o immune cell subpopula ions Raúl Agui e-Gamboa 1† , Niek de Klein 2† , Jenni e di Tommaso 1† , Annique Cla ingbould 2 , Monique GP an de Wijs 2 , Dylan de V ies 2 , Ha m B ugge 2 , Roy Oelen 2 , U mo Võsa 1,3 , Ma ia M. Zo o 1 , Xiaojin Chu 1,4 , Oli ie B. Bakke 1 , Zuzanna Bo ek 1 , Isis Ricaño-Ponce 1 , Pa ick Deelen 2,5 , Cheng-Jiang Xu 4,7 , Mo is Swe z 1,5 , I is Jonke s 1 , Sebo Wi ho 1 , I ma Joos en 6 , Se ena Sanna 1 , Vinod Kuma 1,7 , Hans J. P. M. Koenen 6 , Leo A. B. Joos en 7 , Mihai G. Ne ea 7,8 , Cisca Wijmenga 1 , BIOS Conso ium, Lude F anke 1† and Yang Li 1,4,7*† * Co espondence: Yang.Li@ helmhol z-hzi.de † Raul Agui e-Gamboa, Niek de Klein Jenni e di Tommaso a e con ibu ed equally o his wo k. † Lude F anke and Yang Li a e join ly di ec ed his wo k 1 Depa men o Gene ics, Uni e si y o G oningen, Uni e si y Medical Cen e G oningen, G oningen, he Ne he lands 4 Cen e o Indi idualised In ec ion Medicine (CiiM) & TWINCORE, join en u es be ween he Helmhol z-Cen e o In ec ion Resea ch (HZI) and he Hanno e Medical School (MHH), Feodo -Lynen-S . 7, 30625 Hanno e , Ge many Full lis o au ho in o ma ion is a ailable a he end o he a icle Abs ac Backg ound: Exp ession quan i a i e ai loci (eQTL) s udies a e used o in e p e he unc ion o disease-associa ed gene ic isk ac o s. To da e, mos eQTL analyses ha e been conduc ed in bulk issues, such as whole blood and issue biopsies, which a e likely o mask he cell ype-con ex o he eQTL egula o y e ec s. Al hough his con ex can be in es iga ed by gene a ing ansc ip ional p o iles om pu i ied cell subpopula ions, cu en me hods o do his a e labo -in ensi e and expensi e. We in oduce a new me hod, Decon2, as a amewo k o es ima ing cell p opo ions using exp ession p o iles om bulk blood samples (Decon-cell) ollowed by decon olu ion o cell ype eQTLs (Decon-eQTL). Resul s: The es ima ed cell p opo ions om Decon-cell ag ee wi h expe imen al measu emen s ac oss coho s (R ≥0.77). Using Decon-cell, we could p edic he p opo ions o 34 ci cula ing cell ypes o 3194 samples om a popula ion-based coho . Nex , we iden i ied 16,362 whole-blood eQTLs and decon olu ed cell ype in e ac ion (CTi) eQTLs using he p edic ed cell p opo ions om Decon-cell. CTi eQTLs show excellen allelic di ec ional conco dance wi h eQTL (≥96–100%) and ch oma in ma k QTL (≥87–92%) s udies ha used ei he pu i ied cell subpopula ions o single-cell RNA-seq, ou pe o ming he con en ional in e ac ion e ec . Conclusions: Decon2 p o ides a me hod o de ec cell ype in e ac ion e ec s om bulk blood eQTLs ha is use ul o pinpoin ing he mos ele an cell ype o a gi en complex disease. Decon2 is a ailable as an R package and Ja a applica ion (h ps://gi hub.com/molgenis/sys emsgene ics/ ee/mas e /Decon2) and as a web ool (www.molgenis.o g/decon olu ion). Keywo ds: eQTL, Decon olu ion, Cell ypes, Immune cells © The Au ho (s). 2020 Open Access This a icle is licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License, which pe mi s use, sha ing, adap a ion, dis ibu ion and ep oduc ion in any medium o o ma , as long as you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons licence, and indica e i changes we e made. The images o o he hi d pa y ma e ial in his a icle a e included in he a icle's C ea i e Commons licence, unless indica ed o he wise in a c edi line o he ma e ial. I ma e ial is no included in he a icle's C ea i e Commons licence and you in ended use is no pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly om he copy igh holde . To iew a copy o his licence, isi h p://c ea i ecommons.o g/licenses/by/4.0/. The C ea i e Commons Public Domain Dedica ion wai e (h p://c ea i ecommons.o g/publicdomain/ze o/1.0/) applies o he da a made a ailable in his a icle, unless o he wise s a ed in a c edi line o he da a. Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 h ps://doi.o g/10.1186/s12859-020-03576-5 Backg ound Fo many o he gene ic isk ac o s ha ha e been associa ed o immune diseases by genome-wide associa ion s udies (GWAS), he molecula mechanism leading o disease emains unknown [1]. Mos o hese gene ic isk a ian s a e loca ed in he non-coding egions o he genome, implying ha hey play a ole in gene egula ion [2,3]. Exp es- sion quan i a i e ai locus (eQTL) analysis p o ides a way o cha ac e ize he egula- o y e ec o hese isk ac o s in humans, and many eQTL s udies ha e now been ca ied ou using bulk issues, o example, whole blood [4,5]. Howe e , bulk issues comp ise many di e en cell ypes, and gene egula ion is known o a y ac oss cell ypes [6–8]. In ecen yea s, e o s o desc ibe eQTL e ec s in pu i ied cell subpopula- ions ha e been ca ied ou in speci ic cell ypes [9]. Un o una ely, he leng h and cos o he s udy p o ocols ha e limi ed hese s udies o small sample sizes and only a ew cell ypes. Cu en de elopmen s on single cell (sc) RNASeq echnologies ha e gi en ise o sc-eQTLs, an app oach ha , al hough p omising, is s ill bound o a limi ed numbe o indi iduals, which he eby limi s he numbe o de ec able cell ype in e - ac ion (CTi) eQTLs. Ne e heless, he abili y o pinpoin he CT in which a isk ac o exe s an eQTL e ec could help us o unde s and i s ole in disease. S a is ical app oaches o de ec CT e ec s using issue exp ession p o iles ha e mainly been de eloped o e alua e gene by en i onmen in e ac ion (GxE) e ms, o example o de ec CT eQTLs o myeloid and lymphoid lineages using only whole blood gene exp ession and by e alua ing he in e ac ion be ween geno ype and cell p o- po ions o neu ophils and lymphocy es in whole blood [10]. A second s udy linked eQTL genes o p oxy genes h ough co ela ion; hese p oxy genes we e hen associ- a ed wi h in insic o ex insic ac o s such as cell p opo ions o in lamma ion ma ke s [11]. Howe e , hese e o s ocused on exploi ing only one GxE e m, o on indi ec ly linking he CT p opo ions o gi en eQTL, a he han di ec ly asce aining he in e - ac ion be ween all he main cell p opo ions comp ising he bulk issue and geno ype. Un o una ely, quan i ying cell p opo ions, in pa icula a e subpopula ions ( o al abundance ≤3% in ci cula ing whi e blood cells), is expensi e and ime-consuming. Hence, quan i ying immune cell p opo ions in la ge unc ional genomics coho s is no common p ac ice. He e we p esen and alida e Decon2, a compu a ional and s a is ical amewo k ha can (1) p edic he p opo ions o known ci cula ing immune cell subpopula ions (Decon-cell), and (2) combine hese p edic ed p opo ions wi h whole blood gene ex- p ession and geno ype in o ma ion o assign bulk eQTL e ec s in o CTi eQTLs (Decon-eQTL). Ou wo-s ep amewo k p o ides an imp o emen o e p e iously published me hods. Unlike ea lie me hods [12], Decon-cell does no ely on any p io in o ma ion abou ansc ip ome p o iles om pu i ied cell subpopula ions. I only e- qui es quan i ica ion o he cell p opo ions comp ising he bulk issue, in his case whole blood. Decon-cell iden i ies signa u e genes ha co ela e wi h cell p opo ions in a bulk issue. Secondly, Decon-eQTL is he i s app oach in which all majo cell p opo ions ( he majo cell ypes o which he sum o p opo ions pe sample is ap- p oxima ely 100%) o bulk blood issue a e inco po a ed in o an eQTL model simul an- eously. Decon-eQTL can hen be used o sys ema ically es o any signi ican in e ac ion be ween each CT and geno ype, while also con olling o he e ec on ex- p ession o he o he cell ypes. Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 2 o 23 We gene a ed he Decon-cell p edic i e models using da a om he 500FG coho [13], whe e quan i ica ion o immune cell ypes was ca ied ou using FACS [14] and RNA-Seq-based bulk whole blood ansc ip ome p o iles we e a ailable o 89 samples [15]. By using a c oss- alida ion app oach, we we e able o accu a ely p edic 34 ou o 73 cell sub ypes using only whole blood gene exp ession. Fo alida ion, we applied Decon-cell o h ee independen coho s (Li elines Deep [16], n= 627; Leiden Longe i y coho [17], n= 660 and he Ro e dam S udy [18], n= 773) o which bo h blood RNA-seq and measu ed cell p opo ion da a a e a ailable (neu ophils, lymphocy es and CD14+ monocy es and g anulocy es). Addi ionally, we benchma ked Decon-cell p edic ion pe o mance agains wo o he exis ing me hods ha quan i y immune cell composi ion using gene exp ession p o iles om whole blood on hese h ee independ- en coho s. A e showing ha we can accu a ely p edic ci cula ing immune cell p o- po ions, we applied Decon-cell o es ima e cell p opo ions in 3194 indi iduals om he BIOS coho [16,19–21] o whom bo h whole blood RNA-seq and geno ypes we e a ailable. The BIOS coho is a aluable esou ce o unc ional genomics s udies whe e ex ensi e cha ac e iza ion o he gene ic componen on gene exp ession [11] and epi- gene ics [22] ha e been pe o med. We in eg a ed whole blood exp ession and geno- ype in o ma ion and p edic ed cell p opo ion wi h Decon-eQTL o decon olu e 16, 362 signi ican whole blood cis-eQTLs op e ec s in o CT in e ac ing eQTLs (CTi eQTLs). These decon olu ed CTi eQTL esul s we e comp ehensi ely alida ed using ansc ip ome p o iles om pu i ied cell subpopula ions [23], eQTLs and ch oma in ma k QTLs om pu i ied cell ypes [9] and eQTLs om single-cell expe imen s [24]. We also sys ema ically compa ed he pe o mance o Decon-eQTL agains he mos used me hod [10] ha de ec cell ype eQTL e ec s using whole blood exp ession p o iles. Resul s Decon-cell accu a ely p edic s he p opo ions o known immune cell ypes In o de o assign he cell ypes om which an o e all eQTL e ec om a bulk issue sample (e.g. whole blood) a ise, we need h ee ypes o in o ma ion: geno ype da a, is- sue exp ession da a and cell ype p opo ions (Fig. 1). He e we p opose a compu a- ional me hod ha p edic s he cell p opo ions o known immune cell ypes using gene signa u es in whole blood exp ession da a using a machine-lea ning app oach. Decon-cell employs he egula ized eg ession me hod elas ic ne [26] o de ine se s o signa u e genes o each cell ype. In o he wo ds, hese signa u es we e selec ed as ha ing he bes p edic ion powe o indi idual cell p opo ions. The e a e 89 samples in he 500FG coho wi h bo h whole blood RNA-seq and quan i ica ion o 73 immune cell subpopula ions by FACS. This da a was used o build he p edic ion models o es ima ing cell subpopula ions by Decon-cell. Fi s , we de e - mined which o he 73 cell subpopula ions could be eliably p edic ed by Decon-cell. A wi hin-coho c oss- alida ion s a egy was employed by andomly di iding 89 samples (Fig. 1) in o aining and es se s (70 and 30% o he samples, espec i ely). A e gen- e a ing a model using each aining se , we applied he p edic ion models o each cell ype o he samples in he es se s. We compa ed he p edic ed and measu ed cell p o- po ion o each cell ype using Spea man co ela ion coe icien s o e alua e he Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 3 o 23 p edic ion pe o mance. We epea ed his p ocess 100 imes and hen used he mean o he co ela ion coe icien in all 100 i e a ions o e alua e he p edic ion pe o m- ance. We we e able o p edic 34 ou o 73 cell subpopula ions using whole blood gene exp ession da a a a h eshold o mean R ≥0.5 ac oss all 100 i e a ions (Fig. 2a, Supple- men a y Fig. 1, Supplemen a y Table 1). The numbe o signa u e genes selec ed in ou models o p edic ing cell p opo ions a ied ac oss he cell ypes, anging om 2 o 217 signa u e genes (Supplemen a y Fig. 2A, Supplemen a y Table 1), and hey we e in- dependen o he a e age abundance o hese cell ypes in whole blood (R = 0.02, Spea - man co ela ion coe icien , Supplemen a y Fig. 2A). In pa icula , cell ypes ha a e abundan in whole blood (g anulocy es-neu ophils, CD4+ T-cells and CD14+ mono- cy es) we e p edic ed wi h high con idence (co ela ion be ween p edic ed and mea- su ed alues, R ≥0.73). Rema kably, we we e also able o p edic a numbe o less Fig. 1 Wo k low o applica ion o Decon2 o p edic cell coun s ollowed by decon olu ion o whole blood eQTLs. Using whole blood exp ession and FACS da a o 500FG samples, Decon-cell p edic s cell p opo ions wi h selec ed ma ke genes o ci cula ing immune cell subpopula ions. Valida ions o Decon-cell we e ca ied ou on h ee independen coho s o which measu emen s o neu ophils/g anulocy es, lymphocy es and monocy es CD14+ we e a ailable along wi h exp ession p o iles o whole blood. Benchma king o Decon-cell was pe o med agains CIBERSORT [25] and xCell [12]. Decon-cell was applied o an independen coho (BIOS) o p edic cell coun s using whole blood RNA-seq. Decon-eQTL subsequen ly in eg a es geno ype and issue exp ession da a oge he wi h p edic ed cell p opo ions o samples in BIOS o de ec cell ype eQTLs. We alida ed Decon-eQTL using mul iple independen sou ces, including exp ession p o iles o pu i ied cell subpopula ions, eQTLs and ch oma in ma k QTLs (cmQTLs) om pu i ied neu ophils, monocy es CD14+ and CD4+ T cells [9], and single-cell eQTL esul s [24]. Benchma king o Decon-eQTL was ca ied ou o compa ison wi h a p e iously epo ed me hods ha de ec ed cell ype–eQTL e ec s using whole blood exp ession da a, i.e. he Wes a e al. [10] Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 4 o 23 abundan cell subpopula ions, including NK cells, CD8+ T-cells, non-NK T-cells (CD3- CD56-), CD4+ cen al memo y, CD4+ e ec o memo y T-cells and egula o y T-cells (Supplemen a y Fig. 2A), as de e mined by FACS. Cell ypes wi h a low p edic ion pe - o mance (R < 0.5) a e hose ha ha e ew signa u e genes wi h exp ession le els ha co ela e su icien ly (i.e. absolu e R < 0.3) wi h he measu ed cell p opo ions in whole blood (Supplemen a y Fig. 2B-C). Fo each o he 34 p edic able cell ypes, we used Decon-cell o build models o p edic ing hei cell coun s using all 89 samples om he 500FG coho . These models we e applied o 3194 samples in an independen co- ho (BIOS coho ) o p edic cell p opo ions o ci cula ing immune cell ypes o he subsequen decon olu ion o eQTL e ec s. In addi ion o wi hin-coho alida ion, we es ed ou cell p opo ion models using h ee independen coho s (LLDeep, n= 627; LLS, n= 660; RS, n= 773) in which cell ype abundances we e quan i ied using a Coul e coun e o neu ophils (g anulocy es o RS), lymphocy es and CD14+ monocy es (Fig. 2b, Supplemen a y Fig. 3A-B). In LLDeep, we we e able o accu a ely p edic hese h ee cell ypes wi h Spea man co el- a ion coe icien s o R = 0.73, R = 0.89 and R = 0.73, espec i ely. Fo LLS and RS, he p edic ion pe o mance was simila ly accu a e o neu ophils and lymphocy es (R = 0.76 o neu ophils, R = 0.84 o lymphocy es), bu less so o monocy es (R = 0.50 o CD14+ monocy es and p opo ions in LLS and R = 0.74 o g anulocy es, R = 0.83 o lymphocy es and R = 0.28 o CD14+ monocy es in RS). Fig. 2 P edic ion o cell p opo ions using whole blood ansc ip ome by Decon-cell. aDis ibu ion o p edic ion pe o mance (Spea man co ela ion coe icien ) o he 34 p edic able cell ypes in 100 i e a ions o p edic ion wi hin he 500FG coho . bC oss- coho alida ion in an independen Li elines-Deep coho (n= 627): he measu ed and p edic ed cell p opo ions o neu ophils (gi en by g anulocy es in 500FG), lymphocy es and monocy es a e compa ed Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 5 o 23 Nex , in o de o benchma k Decon-cell, we compa ed i s p edic ion pe o mance agains wo o he exis ing ools ha quan i y he abundance o known immune cell ypes using bulk whole blood exp ession p o iles: CIBERSORT [25] and xCell [12]. We ob ained he p edic ed p opo ions by CIBERSORT and en ichmen sco es o ci cula - ing immune cells by xCell o he samples in h ee di e en coho s: LLDeep, LLS and RS (Supplemen a y Fig. 4A-B). Fo each cell ype, Decon-cell ou pe o ms CIBERSORT and xCell (Supplemen a y Fig. 3B). The sca e plo s o p edic ed s measu ed alues (Supplemen a y Fig. 3A, Supplemen a y Fig. 4A-B) u he demons a e ha he be e pe o mance o Decon-cell is no due o cell p opo ion ou lie s. Finally, we e alua ed whe he he signa u e genes showed CT exp ession in hei ele- an pu i ied cell ypes using BLUEPRINT [23] RNA-seq da a om he pu i ied cell subpopula ions. He e we ocused on cell ypes wi h mo e han h ee samples measu ed, which included neu ophils, CD14+ monocy es, CD4+ T-cells and B-cells. The signa- u e genes showed o e all highe exp ession in hei ele an cell subpopula ions com- pa ed o o he cell subpopula ions. In e es ingly, he signa u e genes we e also able o clus e he samples o he ele an CT using unsupe ised hie a chical clus e ing (Sup- plemen a y Fig. 5A-D). Toge he , ou esul s demons a e ha he gene signa u es iden i ied by Decon-cell using only whole blood gene exp ession da a a e p edic i e o he p opo ions o ci cula ing immune cell subpopula ions. To acili a e he cell p opo ion p edic ion o new samples using whole blood RNA- seq, we ha e made he Decon-cell p edic ion models and gene signa u es a ailable in an R package (Decon-cell) and as a web ool (www.molgenis.o g/decon olu ion). These wo implemen a ions allow use s o p e-p ocess hei RNA-seq exp ession coun s and es ima e cell p opo ions using he p e-es ablished models o 34 cell ypes in whole blood. In addi ion, he Decon-cell R package allows use s o gene a e Decon-cell-like gene signa u es o p edic hei own cell p opo ions, which equi es he inpu o bulk exp ession p o iles and cell p opo ions o gene a e new Decon-cell p edic i e models. Decon-eQTL iden i ies which cell ypes con ibu e o he whole blood eQTL e ec As we know, eQTL analysis using whole blood bulk exp ession da a ails o dis inguish be ween a gene al eQTL p esen in all cell ypes and an e ec mainly ound in a subse o he cell ypes. We he e o e p opose a new app oach, called Decon-eQTL, ha as- signs he o e all bulk eQTL in o CT e ec s. Using he cell p opo ions in whole blood, i is possible o o mally es i he gene ic e ec is in e ac ing wi h he cell p opo ions. Mo e explici ly, we include bo h he geno ype and all majo CT p opo ions o in e es in a linea model, and sys ema ically es i he e is a signi ican in e ac ion e ec be- ween geno ype and each o he cell p opo ions in he a ia ion o gene exp ession in whole blood. A he same ime, he model used by Decon-eQTL con ols o he e ec s o he emaining cell ypes on gene exp ession. In his way, whole blood exp ession da a, geno ypes and (p edic ed) cell p opo ions can be in eg a ed o assign a CTi e ec om a bulk eQTL (Fig. 1). We applied Decon-eQTL o 3198 samples (BIOS coho ) wi h ansc ip ome le els (RNA-seq), geno ype in o ma ion and cell p opo ions p edic ed by Decon-cell. Whole blood cis-eQTL mapping yielded 16,362 whole blood eQTLs ( alse disco e y a e (FDR) ≤0.05). Fo each o hese whole blood cis-eQTLs, we applied Decon-eQTL wi h Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 6 o 23 a ocus on 6 majo cell subpopula ions: g anulocy es, CD14+ monocy es, CD4+ T-cells, CD8+ T-cells, B-cells and NK cells. These cell ypes we e selec ed because he sum o hei ela i e pe cen ages was close o 100% and none o hese cell ype pai s had an absolu e co ela ion coe icien R ≥0.75. Decon-eQTL compu a ionally assigned 4139 CTi eQTLs om hese subpopula ions, e lec ing 3812 genes and 3650 SNPs. 25% o he whole blood eQTLs ha e a signi ican (FDR ≤0.05) CTi eQTL e ec gi en Decon- eQTL. The majo i y (31%) o he o al CTi eQTL e ec s de ec ed we e ound o be as- socia ed o g anulocy e p opo ions, possibly because g anulocy es comp ise ~ 70% o ci cula ing whi e blood cells (Fig. 3a). The majo i y (74%) o CTi eQTLs de ec ed by ou me hod we e assigned o a single cell ype (Supplemen a y Fig. 6A). Simila ly, we ind almos no sha ing be ween cell ypes in single-cell eQTLs om 112 indi iduals. Howe e , i should be no ed ha hese eQTLs a e likely no exclusi ely p esen o his pa icula cell ype in biology, bu ha he s a is ical powe gi en ou sample size was su icien o de ec he in e ac ion e ec s ha we desc ibe as CTi eQTL in his pa icu- la cell ype. Decon-eQTL was only able o ind a ew cases o sha ing o CTi eQTLs be ween cell ypes, likely due o a lack o powe o he in e ac ion model. An example Fig. 3 Decon olu ion o whole blood eQTLs in o CTi eQTLs. Decon-eQTL de ec s CTi eQTLs by in eg a ing p opo ions o cell subpopula ions (p edic ed by Decon-cell), gene exp ession and geno ype in o ma ion. a Numbe o decon olu ed CTi eQTLs in each cell ype using whole blood RNA-seq da a o 3189 samples in BIOS coho . bDis ibu ion o Spea man co ela ion coe icien s be ween exp ession le els o CTi eQTL genes and cell coun s o each cell subpopula ion. The CTi eQTL genes show posi i e and s a is ically highe co ela ion (Spea man) wi h he ele an cell ype p opo ions as compa ed o he es (T- es p- alue < 0.05) in an independen coho (500FG) Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 7 o 23 o such a sha ed CTi eQTLs can be seen o he NOD2 gene, whe e Decon-eQTL de- ec ed a s ong g anulocy e-eQTL e ec alongside a smalle opposi e e ec in CD14+ monocy es. This opposi e e ec has also been p e iously desc ibed in eQTL s udies on pu i ied CD14+ monocy es and neu ophils [8]. These esul s demons a e ha he e - ec s o cell p opo ions on gene exp ession should be aken in o accoun when in e - p e ing eQTLs de i ed om bulk issues. Decon-eQTL p io i izes genes o ele an cell ypes CTi eQTL genes a e expec ed o ha e highe exp ession le els in hei ele an cell ypes, and hei exp ession in whole blood should he e o e be co ela ed wi h he p opo ions o hese ele an cell ypes. To es his, we e alua ed i he exp ession le els o he CTi eQTL genes de ec ed in heBIOScoho we eco ela edwi h hei ele an cell p opo ions, and compa ed his o he co ela ion wi h non- ele an cell ypes. We calcula ed he Spea man co ela ion coe icien s be ween he exp ession o he iden i ied CTi eQTL genes and he measu ed cell p opo ions in he 500FG coho (n= 89). We hen compa ed he co ela ion coe icien s we ob ained he e wi h hose be ween exp ession and he emaining cell p opo ions. Fo each o he six cell subpopula ions we e alua ed in Decon-eQTL, hei CTi eQTL genes had a signi ican ly highe co ela ion wi h hei ele an cell subpopu- la ion han wi h o he cell ypes (T- es , p- alue < 0.05) (Fig. 3b). As such, his e- sul shows a signi ican associa ion be ween CTi eQTL genes and he p opo ion o hei ele an CT in an independen coho . Nex , we e alua ed whe he he signi ican CTi eQTL genes we e o e -exp essed in hei ele an cell subpopula ion compa ed o eQTL genes ha we e ound o be non- signi ican CTi eQTLs o he same cell ype. Fo his pu pose, we made use o he pu i- ied neu ophil, CD14+ monocy e, CD4+ T-cell and B-cell RNA-seq da a om he BLUE- PRINT da ase . We include hese cell ypes because hey we e he only ones wi h mo e han h ee samples measu ed. Fo each o he ou cell ypes, we obse ed ha he exp es- sion o CT eQTL genes de ec ed by Decon-eQTL was signi ican ly highe (T- es , p- alue ≤0.05) han he exp ession o non-signi ican Decon-eQTL genes (Fig. 4a). We also ob- se ed ha he decon olu ed eQTL genes om g anulocy es showed a ela i ely wide ange o a ia ion han he CT eQTL genes om he o he h ee subpopula ions. We hy- po hesized ha his could be explained by he ac ha g anulocy es comp ise ~ 70% o he cell composi ion in whole blood, hus gi ing us he powe o de ec eQTL o lowly exp essed genes in g anulocy es. This is pa ly suppo ed by he obse a ion ha he a i- a ion o exp ession in whole blood o g anulocy e CTi eQTL genes was signi ican ly g ea e han o hose CTi eQTL genes decon olu ed o he o he i e cell subpopula ions (F- es , p- alue ≤0.05, Supplemen a y Fig. 7). Fu he mo e, by using publicly a ailable ansc ip ome p o iles (GSE78840 [27]) o pu i ied NK cells and CD4+ T cells, we assessed i he di e en ially exp essed genes ac oss he wo cell ypes we e en iched o eGenes o decon olu ed CT eQTLs. He e we obse ed ha he CD4+ di e en ially exp essed genes (Adjus ed P- alue ≤0.05) we e signi ican ly en iched o CD4+ T cell eQTLs (Fishe exac P= 1.8 × 10 −17 ), whe eas NK cell di e en ial genes (Adjus ed P- alue ≤0.05) we e signi ican ly en iched o NK cell eQTLs (Fishe exac P= 2.3 × 10 −18 ) as shown in Fig. 4b. Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 8 o 23 In summa y, we we e able o show ha he eQTL genes de ec ed by Decon-eQTL a e ansc ip ionally ac i e in hei ele an cell ype because ha is whe e hey a e mo e highly exp essed. CT eQTLs iden i ied by Decon-eQTL in whole blood a e eplica ed in pu i ied cell eQTL da ase s To alida e he CT eQTLs de ined by Decon-eQTL, we u ilized eQTLs iden i ied om pu i ied neu ophils, CD4+ T-cells and CD14+ monocy es [9]. We i s compa ed he absolu e e ec sizes o eQTLs om pu i ied cells ha a e also signi ican ly decon o- lu ed CTi eQTLs o he e ec sizes o eQTLs om pu i ied cells ha a e also non- signi ican decon olu ed CTi eQTLs o his cell ype. Fo all h ee cell popula ions, e - ec sizes in ou decon olu ed CTi eQTLs we e signi ican ly highe han he e ec sizes o eQTLs wi hou a signi ican CTi eQTL (Wilcoxon es , p- alue ≤0.05, Fig. 4c). Nex , we assessed he speci ici y o ou decon olu ed CTi eQTLs by e alua ing CTi eQTL e - ec sizes in non- ele an cell subpopula ions. Fo example, we compa ed he e ec sizes o decon olu ed g anulocy e CTi eQTLs agains hose wi h non-signi ican decon- olu ed g anulocy e CTi eQTLs using he e ec sizes o pu i ied CD4+ T-cell eQTLs. No ably, we obse ed no s a is ically signi ican di e ences using e ec sizes om non- ele an cell subpopula ions (see o -diagonal compa isons in Supplemen a y Fig. 8), which u he suppo s he biological ele ance o ou decon olu ed CTi eQTLs. How- e e , when compa ing he e ec sizes in he pu i ied eQTLs o only he CTi eQTLs ha Fig. 4 Valida ion o CTi eQTLs. aThe exp ession o CTi eQTL genes in pu i ied cell subpopula ions om BLUEPRINT [23] a e signi ican ly highe in he ele an cell subpopula ion when compa ed o o he a ailable cell sub ypes (g een o g anulocy e eQTL genes showing exp ession o pu i ied neu ophils; o ange o monocy es; pu ple o CD4+ T cells; pink o B cells). bGenes di e en ially exp essed (Adjus ed p- alue ≤0.5) be ween CD4+ T cells and NK cells a e signi ican ly en iched o CT eQTLs e ec s on CD4+ T cells (do s in pu ple, Fishe exac P= 1.8 × 10 17 ) and NK Cells (do s in yellow, Fishe exac P= 2.3 × 10 18 ), espec i ely. cCTi-eQTLs (FDR ≤0.05) show signi ican ly la ge e ec sizes in he pu i ied cell eQTL da a [9] compa ed o he es o he whole blood eQTLs o which we do no de ec a cell ype e ec , as shown o decon olu ed g anulocy e eQTLs in neu ophil-de i ed eQTLs (g een),monocy es (o ange) and CD4+ T cells (pu ple) Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 9 o 23 geno ypes and exp ession da a is a ailable and could po en ially aid in unde s anding he molecula e ec s o gene ic isk ac o s associa ed wi h complex diseases a he cell-subpopula ion le el. Ou me hod makes i possible o c ea e CT gene egula o y ne wo ks ha could explain he di e en e ec s ha each CT has on a complex disease in a cos -e icien way. Since Decon2 only equi es gene exp ession and geno ype in o - ma ion o decon olu e bulk blood eQTLs in o CTi eQTLs, i is possible o e-analyze exis ing bulk blood RNA-seq da a o which geno ypes a e also a ailable. In his sce- na io, we would use Decon-cell o p edic cell p opo ions in whole blood and ob ain CT in o ma ion on many mo e eQTLs h ough an inc ease in sample size. In addi ion o whole blood, he me hods behind Decon2 can po en ially be gene alized o use an- sc ip ional p o iles de i ed om any o he ype o bulk issue, such as biopsies om u- mo s o o he solid issues implica ed in complex disease e iology. Howe e , he me hod has no ye been es ed in o he issues. Ou me hods can hence aid in he de- ec ion o gene ic e ec s on gene exp ession in a e cell subpopula ions in bulk issues. Me hods RNA-seq da a collec ion in 500FG coho We selec ed a ep esen a i e subse o 89 samples om he 500 pa icipan s o he 500FG coho , which is pa o he Human Func ional Genomics P ojec (HFGP). Ou subse was balanced o age and sex based on he o iginal dis ibu ion in he coho . RNA was isola ed om whole blood and globin ansc ip s we e subsequen ly il e ed by applying he Ambion GLOBINclea ki . The samples we e hen p ocessed o se- quencing using he Illumina T uSeq 2.0 lib a y p epa a ion ki . Pai ed-end sequencing o 2 × 50-bp eads was pe o med on he Illumina HiSeq 2000 pla o m. The quali y o he aw eads was checked using Fas QC (h p://www.bioin o ma ics.bab aham.ac.uk/ p ojec s/ as qc/). Read alignmen was pe o med wi h STAR 2.3.0 [32,33] using he hu- man Ensembl GRCh37.75 as e e ence, and he aligned eads we e so ed using SAM- Tools [34]. Las ly, gene-le el quan i ica ion o he eads was done using HTSeq [35]. RNA-seq p epa a ion and da a p ocessing in he BIOS coho RNA was isola ed om whole blood and globin ansc ip s we e subsequen ly il e ed by applying he Ambion GLOBINclea ki . Lib a y p epa a ion was pe o med using he Illumina T uSeq 2 lib a y p epa a ion ki . Nex , Illumina HiSeq 2000 was used o pai ed-end sequencing o 2 × 50 bp eads while pooling 10 samples pe lane and expec - ing > 15 million ead pai s pe sample. Read se s we e gene a ed using CASAVA, e aining only hose eads ha passed Illumina Chas i y Fil e . Quali y con ol o he eads was e alua ed using Fas QC (h p://www.bioin o ma ics. bab aham.ac.uk/p ojec s/ as qc/). Adap o sequences we e immed ou using cu adap ( 1.1) wi h de aul se ings. Low quali y ends o eads we e emo ed using Sickle ( 1.200) (h ps://gi hub.com/najoshi/sickle). Reads we e hen aligned using STAR 2.3.0e [33]. All SNPs p esen in he Genome o he Ne he lands (GoNL) wi h MAF ≥0.01 we e masked om he eads o a oid e e - ence mapping bias. Read pai s wi h a mos eigh misma ches and mapping o a mos i e posi ions we e used. Quan i ica ion o coun s pe genes was done using Ensembl .71 anno a ion (which co esponds o GENCODE .16). Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 16 o 23 Geno ype da a o he BIOS coho Geno ype in o ma ion was independen ly gene a ed o each o he coho s, u he de- ails on da a collec ion and me hods used o geno yping can be ound in hei pape s (CODAM [36], LLDeep [16], LLS [17], RS [18] and NTR [21]). Geno ypes we e ha monized o GoNL wi h Geno ype Ha monize [37] and impu ed wi h IMPUTE2 [38] using GoNL as e e ence panel. SNPs wi h an impu a ion sco e below 0.5, a Ha dy-Weinbe g equilib ium P- alue smalle han 1 × 10 −4 , a call a e below 95%, o a MAF smalle han 0.05 we e il e ed ou . Fo u he analysis, only eSNPs om whole blood cis-eQTL op e ec s we e subsequen ly used in Decon-eQTL. Quan i ica ion o cell p opo ions in 500FG coho Inclusion c i e ia and u he desc ip ion o he pa icipan s o he 500FG coho can be ound a h p://www.human unc ionalgenomics.o g. A o al o 73 manually anno- a ed immune cell subpopula ions we e quan i ied using 10-colo low cy ome y. To minimize biological a iabili y, cells we e p ocessed immedia ely a e blood sampling and ypically analyzed wi hin 2–3 h. Cell popula ions we e ga ed manually as p e iously desc ibed [14]. Cis-eQTLs in he BIOS coho Fo cis-QTL mapping, we es ed o associa ion be ween genes and SNPs loca ed wi hin 250 kb o a gene cen e . SNPs wi h MAF ≥0.01, call a e = 1 and Ha dy-Weinbe g equi- lib ium p- alue ≥0.0001 we e included. eQTLs we e decla ed o be signi ican a FDR < 0.05. P e-p ocessing o RNA-seq and QTL mapping was pe o med using a cus om eQTL pipeline ha has been desc ibed p e iously [11]. No maliza ion and co ec ion o gene exp ession da a o decon olu ion o eQTL e ec s To al ead coun s om HTSeq we e i s no malized using he immed means o M (TMM) alues32. TMM exp ession alues we e hen log2 ans o med. Fo p edic ing cell p opo ions, we used scaled exp ession da a in bo h he 500FG and BIOS coho s. Fo he decon olu ion o eQTLs, he exp ession was log2 ans o med and co ec ed o he e ec s o coho , age, sex, GC con en , RNA deg ada ion a es, lib a y size and numbe o de ec ed genes pe sample using a linea model. The co ec ed exp ession da a was hen exponen ia ed o main ain he o iginal linea ela ionship ac oss ead coun s (gene exp ession) and cell p opo ions. Gene al desc ip ion o Decon2 Decon2 is a s a is ical amewo k o es ima ing cell coun s using molecula p o iling such as exp ession da a om he e ogeneous samples (Decon-cell) and consecu i e de- con olu ion o exp ession quan i a i e ai loci (Decon-eQTL) in o each cell subpopu- la ion. To p edic cell p opo ion le els using Decon-cell buil in models, i ’s only inpu is a ma ix As inpu Decon-cell akes a able o no malized gene exp ession coun s, wi h samples as columns and genes as ows, and ou pu s a able o p edic ed cell coun p opo ions o cell ypes ha we e included in he aining model.. Decon-cell also en- ables he use o gene a e i s own cus om models, o which i equi es a ma ix o gene exp ession o ain he model and a ma ix o measu ed cell p opo ions; his will Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 17 o 23 ou pu a lis wi h one speci ic model o each o he cell ypes included. A ma ix able o no malized gene exp ession le elscoun s, a ma ix able o p edic ed o measu ed cell coun p opo ions, and a ma ix able o geno ype dosages (0 o homozygous e e ence, 1 o he e ozygous, and 2 o homozygous al e na i e), las lyand a able wi h he SNP + gene combina ions o es , a e used as inpu o Decon-eQTL, and his ou pu s o each SNP + gene combina ion he be a and p- alue o he cell- ype dependen eQTL e ec . See supplemen al Fig. 20 o a g aphical o e iew. P edic ion o cell p opo ions using gene exp ession le els om bulk issue (Decon-cell) Fo cell coun p edic ion, exp ession da a is TMM no malized, log2(exp ession+ 1) ans o med and z- ans o med (scaled). We p oposed ha he abundance o molecula ma ke s such as gene exp ession could be used as p oxies o p edic cell p opo ions. This can be ep esen ed as: Ckj ¼βki Yij þekj ð1Þ whe e exp ession da a is Y ij o genes i=1, 2,…, G and samples j=1, 2, …,Nand cell coun da a is C kj o sample jin cell ype k(k = 1, 2, …, K). β ki ep esen s he coe i- cien s o gene iin de e mining cell coun s o cell ype ko a complex issue. e kj is he e o e m. In o de o selec only he mos in o ma i e genes o p edic ing cell coun s, we im- plemen ed a ea u e selec ion scheme by applying an elas ic ne (EN) egula ized e- g ession [26]. In he EN algo i hm, he β k Ya e es ima ed by minimizing: Ck−βkY    2subjec o 1−αðÞβk    2þαβ k   1≤sð2Þ sis a uning pa ame e ha limi s he numbe o ea u es ha will be included in he inal p edic o model. We es ima e he bes spe cell ype by applying a 10- old c oss- alida ion app oach, whe e he mos op imal penal y pa ame e (α) was ob ained. Decon olu ion o eQTL e ec s (Decon-eQTL) Decon-eQTL models he exp ession le el in he bulk issue by conside ing he gene ic con ibu ion o mul iple cell ypes p esen in he sys em. Fo iden i ying he CT eQTL e ec , he in e ac ion e m be ween a pa icula cell ype and geno ype was es ed o s a is ically signi ican con ibu ion o he explained a iance on he exp ession le els o pa icula genes, while accoun ing o he emaining cell p opo ions. I we conside a gene ic eQTL linea model o whole blood i can be desc ibed as: y¼aþβ:gþeð3Þ whe e yis he measu ed gene exp ession, a he modeled non-gene ic dependen ex- p ession, g he geno ype coded as 0, 1 o 2, β.g he geno ype-dependen exp ession and e he e o , e.g. unknown en i onmen al e ec s. He e, all h ee e ms a e modeling he e ec o he mix u e o di e en cell ypes p esen in blood. In an RNA-seq-based gene exp ession quan i ica ion o a bulk issue, one could exp ess gene exp ession le els (y) as he sum o coun s (ψ) pe Kcell ypes: Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 18 o 23 y¼XK k¼1ψkð4Þ Fo e e y cell ype, he exp ession le el can be w i en as a gene ic eQTL model (eq. 3) weigh ed by he cell p opo ions. ψ k is a combina ion o he gene ic and non-gene ic con- ibu ion o he cell ype o y. The non-gene ic con ibu ion pe cell ype is β.c,whe ecis he cell coun p opo ions. The gene ic con ibu ion is β k .g:c k .Fo kcell ypes he ex- p ession is hen: y¼XK k¼1ψk¼Σk:βk:ck  þΣk:γk:gck  þeð5Þ whe e yis he measu ed exp ession le els, kis he o al numbe o cell ypes, c k is he cell coun p opo ions o cell ype k,gis he geno ype and eis he e o e m. Since we a e assuming a linea ela ionship be ween o al gene exp ession and he le els o ex- p ession gene a ed by each o he cell ypes composing a bulk issue, he cell p opo - ions a e scaled o sum o 100% such ha he sum o he e ec o he cell ypes equals he e ec in whole blood. He e we assume ha he ue sum o he cell coun s should be e y close o 100% o he o al PBMC coun , which is why we include he 6 cell ypes ha oge he o m he op hie a chy gi en he ga ing s a egy used o quan i y he cell subpopula ions [14]. The geno ype main e ec is no included in he model be- cause he sum o he geno ype e ec pe cell ype should app oxima e he main e ec . Because he con ibu ion o each o he cell ypes o exp ession le el ycanno be nega i e, we cons ain he e ms o he model o be posi i e using Non-Nega i e Leas Squa es [39,40] o i he pa ame e s o he measu ed exp ession le els. Howe e , i he allele ha has a nega i e e ec on gene exp ession is coded as 2, he bes i would ha e a nega i e in e ac ion e m, which would be se o 0. To add ess his, we wan he allele ha causes a posi i e e ec on gene exp ession o always be coded as 2. Howe e , he e ec o an allele can be di e en pe cell ype, he e o e he coding o he SNP should also be di e en pe cell ype. We he e o e un he model mul iple imes, swap- ping he geno ype encoding o one o he in e ac ion e ms each ime. The encoding ha gi es he lowes R-squa ed is hen chosen as he op imal geno ype encoding. Fo he encoding, we limi he numbe o geno ypes ha ha e an opposi e geno ypic encod- ing o a maximum o one in e ac ion e m, as we ha e obse ed ha his leads o no signi ican di e ence when compa ed o using all possible con igu a ions and limi s he numbe o models ha ha e o be un om k 2 o (2*k) + 2. To es i he e is a CT in e ac ion e ec , we un he linea model o eq. 5and, o each CT, un he same model wi h he cell p opo ion:geno ype in e ac ion e m e- mo ed. Fo example, when es ing wo cell ypes he ull model is: y¼β1:c1þβ2:c2þγ1:gc1þγ2:gc2þeð6Þ and he wo models wi h he in e ac ion e ms emo ed a e: y¼β1:c1þβ2:c2þγ1:gc1þe y¼β1:c1þβ2:c2þγ2:gc2þeð7Þ Fo bo h he ull model and he CT models, we calcula ed he sum o squa es using he di e en geno ype con igu a ions de ailed abo e. Fo bo h he ull and he CT models, we hen selec ed he geno ype con igu a ion wi h lowes sum o squa es. Then, Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 19 o 23 o each CT, we es ed i he ull model could signi ican ly explain mo e a iance han he CT model using an ANOVA. We hen applied ou s a egy o 16,362 signi ican whole blood cis-eQTL op e ec s de ec ed using he BIOS coho . We hen co ec he p- alues o mul iple es ing using FDR o each o he cell ypes, i.e. g anulocy e eQTL p- alues we e co ec ed o 16, 362 es s in he same way as CD4+ T cells eQTL p- alues we e co ec ed o he exac same numbe o es s. Wes a e al. in e ac ion model In he Wes a e al. model, exp ession da a is no malized in he same way as in Decon- eQTL. The e ec o he cell ype is p edic ed using a geno ype * cell coun in e ac ion e m: y¼Iþβ1:Gþβ2:cþβ3:cxGþe whe e y is exp ession, I he in e cep , G he geno ype, c he cell coun and c x G he cell coun x geno ype in e ac ion e m. Addi ional es ic ions a e se on he p- alues. Fo neu ophils, i ( he βo he neu ophil xGin e ac ion e m) * ( he βo he Gin e ac ion e m) < 0, he p- alue is se o 1. Fo CD4+ and monocy es, i ( he βo he neu ophil xG in e ac ion e m) * ( he βo he Gin e ac ion e m) > 0, he p- alue is se o 1. Compa ison be ween allelic conco dance Fo he compa ison be ween allelic conco dances, we coun ed he conco dan and dis- co dan eQTLs o each o he cell ype compa isons and did a Fishe exac es be- ween each o he g oups. The p- alues a e Bon e oni-co ec ed. Single-cell eQTLs The sc-eQTLs we e ob ained o 112 indi iduals in he same way as desc ibed in Van de Wijs e al. [24] Fo he allelic di ec ion compa ison, we used all signi ican eQTLs. Classical monocy e and non-classical monocy e eQTLs we e combined and join ly compa ed o Decon-eQTL Monocy es. Supplemen a y in o ma ion Supplemen a y in o ma ion accompanies his pape a h ps://doi.o g/10.1186/s12859-020-03576-5. Addi ional ile 1 : Supplemen a y Figu e 1: P edic ion pe o mance o Decon-cell wi hin 500FG: The Y-axis ep- esen s he 73 immune cell ypes quan i ied by FACS in he 500FG coho . The ba plo on he le panel shows he mean P edic ion Pe o mance (Spea man co ela ion coe icien be ween p edic ed and measu ed cells ac oss 100- old c oss alida ions). On he igh panel, box plo s ep esen he dis ibu ion o he P edic ion Pe o mance wi hin 100 i e a ions o he c oss alida ions. A cu o o mean P edic ion Pe o mance ≥0.5 was applied o de ine p edic able cell ypes (g een). Supplemen a y Figu e 2. Signa u e genes selec ed o p edic ion o cell p opo - ions by Decon-cell: (A) To al numbe o ma ke genes (genes selec ed in ≥80% o all models in he 100 i e a ions) pe p edic able cell ype. Di e en colo s indica e di e en subpopula ions. (B) The numbe o genes signi ican ly co ela ed wi h cell coun s (Spea man co ela ion, adjus ed P≤0.05) (y-axis) shows he o al numbe o signi ican ly co ela ed genes, while he x-axis shows he p edic ion pe o mance (x-axis). (C) Dis ibu ions o he o al numbe o “s ongly”co ela ed genes (absolu e Spea man co ela ion ≥0.3) be ween p edic able and unp edic able cell subpopula ions. Supplemen a y Figu e 3. Compa ison o p edic ion pe o mance be ween Decon-cell and o he exis ing me hods. (A) Pe o mance o Decon-cell: he measu ed (x axis) and p edic ed cell p opo ions (y-axis) we e compa ed o neu ophils (gi en by g anulocy es in 500FG), lymphocy es and monocy es CD14+ and g anulocy es in h ee independen coho s (shown by ow, om op o bo om: LLDeep (n= 627); LLS (n= 660); RS (n= 773)). (B) Compa ison o p edic ion pe o mance o Decon-cell, CIBERSORT and xCell in h ee independen coho s o a o al o 4 majo immune subpopula ions. Supplemen a y Figu e 4. P edic ion pe o mance o xCell and CIBER- SORT in h ee independen Du ch popula ions (LLDeep, n = 627; LLS, n = 660; RS, n = 773).(A) Sca e plo s show- ing he measu ed cell p opo ions o ci cula ing immune cells on he x-axis and he xCell en ichmen sco e on he y-axis. (B) Sca e plo s showing he measu ed cell p opo ions o ci cula ing immune cells on he x-axis and he p edic ed cell p opo ions gi en by CIBERSORT) on he y-axis. Supplemen a y Figu e 5. Exp ession o ma ke Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 20 o 23 genes selec ed by Decon-cell. Exp ession le els (scaled, log2(TPM + 1)) o signa u e genes in he da a in h ee pu i- ied cell subpopula ions: CD4+ T cells (A), neu ophils/g anulocy es (B) and monocy es (C) in he da a om BLUE- PRINT. Cell subpopula ions a e indica ed in di e en colo s by columns. Co ela ion o each o he signa u e genes and he cell subpopula ion pe cen age in he 500FG coho is shown on by he g een ba a he le -hand side o hea map igu e, i.e. da ke g een co esponds o highe co ela ions. Supplemen a y Figu e 6. Many o he CTi eQTL a e cell ype exclusi e. Colo ed ba plo on he le shows he o al numbe o signi ican CTi eQTLs in whole blood eQTLs (as also shown in Fig. 2a). G ay ba plo shows he o al numbe o eQTLs sha ed ac oss he possible combina ions o he six cell subpopula ions unde s udy. Supplemen a y Figu e 7. Va ia ion o gene exp ession ac oss samples o decon olu ed cell- ype eQTLs genes in whole blood. G anulocy e eQTL genes show signi ican ly highe a iance ac oss he BIOS samples (F es p- alue ≤0.05) compa ed o hose om monocy es, CD4+ T cells, CD8+ T cells, B cells and NK cells. Supplemen a y Figu e 8. Valida ion o CTi eQTLs using e ec sizes o eQTLs om pu i ied cells. CTi eQTLs (FDR ≤0.05) om he BIOS coho show a signi ican ly bigge e ec size in pu i ied cell eQTLs [9] om hei ele an cell sub ype as compa ed o o he whole blood eQTLs (diagonal boxed compa i- sons). The o -diagonal compa isons show ha hese eQTL genes a e speci ic o a cell subpopula ion because he di e ences in e ec sizes a e non-signi ican in all bu one case (CD4+ T cell eQTL genes in monocy e-de i ed eQTLs). Supplemen a y Figu e 9. Valida ion o CTi eQTLs using e ec sizes o K27AC QTLs om pu i ied cells. CTi eQTLs (FDR ≤0.05) show a signi ican ly bigge e ec size o K27AC QTLs ha ha e peaks loca ed in he p omo e egion o he eGenes om hei ele an cell sub ype compa ed o he es o he signi ican whole blood eQTLs (diagonal boxed compa isons). The o -diagonal compa isons show ha hese eQTL genes a e speci ic o a cell sub- ype because he di e ences in e ec sizes a e non-signi ican in all bu he compa isons ac oss Neu ophils and Monocy es (CD14+). Supplemen a y Figu e 10. Valida ion o CTi eQTLs using e ec sizes o K4ME1 QTLs om pu i ied cells. CTi eQTLs (FDR ≤0.05) show a signi ican ly bigge e ec size o K4ME1 QTLs (whe e he eGenes is he closes gene agging he K4ME1 QTLs peak) om hei ele an cell sub ype compa ed o he es o he signi i- can whole blood eQTLs (diagonal boxed compa isons). The o -diagonal compa isons show ha hese eQTL genes a e speci ic o a cell sub ype because he di e ences in e ec sizes a e non-signi ican in all bu he compa isons be ween neu ophils and monocy es (CD14+). Supplemen a y Figu e 11. Valida ion o CTi eQTLs using allelic conco dance wi h eQTLs esul s om pu i ied cells. CTi eQTLs (FDR ≤0.05) show high allelic conco dance wi h hei espec i e pu i ied cell eQTLs. Top ow shows allelic conco dance o decon olu ed g anulocy e eQTLs (all in g een) agains neu ophils, monocy es and CD4+ T cells. Second ow shows decon olu ed monocy e eQTLs agains pu i- ied cell eQTLs in he same o de as he op ow. Bo om ow shows he same compa isons as o decon olu ed CD4+ eQTLs. Allelic conco dance o he o -diagonal (compa ing CTi eQLTs wi h non- ele an cell ypes) show a consis en dec ease in allelic conco dance. P- alues a e Bon e oni-co ec ed Fishe exac es s be ween g oups. Supplemen a y Figu e 12. Valida ion o CTi eQTLs using allelic conco dance wi h K27AC esul s om pu i ied cells. CTi eQTLs (FDR ≤0.05) show a high allelic conco dance in hei espec i e pu i ied cell K27AC QTLs. Top ow shows allelic conco dance o decon olu ed g anulocy e eQTLs (all in g een) agains neu ophils, monocy es and CD4+ T cells de i ed om K27AC QTLs. Second ow shows decon olu ed monocy e eQTLs (all in o ange) agains pu i ied cell K27AC QTLs in he same o de as op ow. Bo om ow shows he same compa isons as o decon o- lu ed CD4+ eQTLs (all in pu ple). Allelic conco dance o he o -diagonal (compa ing decon olu ed eQTLs wi h non- ele an cell ypes) show a consis en dec ease in allelic conco dance when compa ed o he ele an cell ype compa isons. P- alues a e Bon e oni-co ec ed Fishe exac es s be ween g oups. Supplemen a y Figu e 13. Al- lelic conco dance be ween whole blood eQTLs and K27AC QTLs o pu i ied neu ophils, CD14+ monocy es and CD4+ T cells. Supplemen a y Figu e 14. Compa ison o whole blood eQTLs wi h eQTLs om single cell RNA-seq Whole blood eQTLs show 89% allelic conco dance o signi ican eQTLs de i ed om scRNA-seq da a, comp ising monocy es CD14+, B cells, CD4+ T cells, CD8+ T cells and NK cells. Supplemen a y Figu e 15 Valida ion o cell ype eQTLs de ec ed in he BIOS coho using he Wes a e al me hod: (A) Exp ession o eGenes in pu i ied cell subpopula ions om BLUEPRINT (g een o g anulocy e eQTL genes showing exp ession o pu i ied neu ophils; o ange o monocy es; pu ple o CD4+ T cells; pink o B cells). (B) CT eQTLs de ec ed by he Wes a me hod show a signi ican ly la ge e ec size in pu i ied cell eQTLs [11] as compa ed o he es o he whole blood eQTLs. Boxed-diagonal shows he compa isons wi h ele an cell ypes whe e he e ec di e ences a e s onge . Supple- men a y Figu e 16 Allelic conco dance a es o cell ype eQTLs de ec ed using he Wes a e al me hod and eQTLs om pu i ied cells. Top ow shows allelic conco dance o g anulocy e CT eQTLs agains neu ophils, mono- cy es and CD4+ T cells. Second ow shows CT monocy e eQTLs agains pu i ied cell eQTLs in he same o de as op ow. Bo om ow shows he same compa isons o CT CD4+ eQTLs. Supplemen a y Figu e 17 Compa ison o Decon-eQTL wi h Wes a e al me hod. O e lap o CT eQTLs de ec ed wi h Decon-eQTL and he Wes a e al me hod and hose ound o be signi ican in pu i ied cell subpopula ions o g anulocy e QTLs (A), CD4+ T cells (B), and monocy es (C). Supplemen a y Figu e 18 Dis ibu ion and co ela ion among ci cula ing cell p opo ions. (A) Sca e plo s show he co ela ions be ween di e en cell subpopula ions in 89 samples om 500FG. Blue line indica es a i ed linea model. Diagonal plo s depic he o e all densi y dis ibu ion pe cell ype. Uppe igh i- angle shows he Pea son co ela ion coe icien o each pai wise compa ison. (B) Co ela ions be ween di e en cell subpopula ions in he BIOS coho ob ained by p edic ion using Decon-cell. Supplemen a y Figu e 19.Gen- e al o e iew o he Decon2 me hod. (A) Gene exp ession can be used o p edic cell coun pe cen ages o cell coun s ha a e al eady ained in he Decon-Cell model. Addi ionally, he model can be ained on di e en cell ypes i exp ession da a and cell coun p opo ions a e a ailable. (B) Decon-eQTL models he cell ype dependen eQTL e ec using exp ession, geno ype, and measu ed cell coun p opo ions o , i una ailable, p edic ed cell coun p opo ions. Addi ional ile 2 : Supplemen a y Table 1: Ensembl IDs and symbol names o he ma ke genes selec ed by Decon-cell o he 34 p edic able ci cula ing immune cell p opo ions. Addi ional ile 3 : Supplemen a y Table 2: Summa y s a is ics om Decon-eQTLs o he 16,362 whole blood eQTLs. Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 21 o 23 Abb e ia ions eQTL: exp ession quan i a i e ai loci; CT: Cell ype; CTi: Cell ype in e ac ion; GWAS: Genome-wide associa ion s udies; sc: single cell; GxE: Gene by en i onmen in e ac ion; TMM: T immed means o M Acknowledgemen s We hank K Mc In y e and J Senio o edi ing he inal ex . We hank T. Spenkelink o he DeconCell web ool design. We hank he UMCG Genomics Coo dina ion cen e , he UG Cen e o In o ma ion Technology and hei sponso s BBMRI-NL & Ta Ge o s o age and compu e in as uc u e, and he Cen e o In o ma ion Technology o he Uni e si y o G oningen o hei suppo and o p o iding access o he Pe eg ine high pe o mance compu ing clus e . Au ho s’con ibu ions C.W., L.F. and YL ini ialized he s udy. Y.L. and L.F. di ec ed and supe ised he p ojec . Y.L. de eloped he s a is ical amewo k, oge he wi h L.F.. R.A-G, N.K., L.F., and Y.L., pe o med da a analysis and in e p e a ion. J.D.T. was in ol ed in he ini ial analysis. N.K. and R.A-G. made he so wa e and web ool. A. C, U.V., M. Z, X.C., O.B.B., Z.B., I.R.P., P.D., C.J.X., M.S., I.J. [1], S.W., I.J. [2], S.S., V.K., H.J.P.M.K., L.A.B.J., M.G.N., M.W., D.V., H.B., R.O. and C.W. con ibu ed o da a collec ion, da a analysis and in e p e a ion. R.A-G, N.K., L.F., and Y.L. d a and e ise he manusc ip . All au ho s ha e ead and app o ed he manusc ip . Funding L.F. is suppo ed by g an s om he Du ch Resea ch Council (ZonMW-VIDI 917.164.455 o M.S. and ZonMW-VIDI 917.14.374 o L.F.), and by an ERC S a ing G an , g an ag eemen 637640 (ImmRisk). Y.L. was suppo ed by an ZonMW-O Road g an (91215206). The HFGP is suppo ed by a Eu opean Resea ch Council (ERC) Consolida o g an (ERC 310372). This s udy was u he suppo ed by an IN-CONTROL CVON g an (CVON2012–03) and a Ne he lands O ganiza ion o Scien i ic Resea ch (NWO) Spinoza p ize (NWO SPI 94–212) o M.G.N.; an ERC ad anced g an (FP/ 2007–2013/ERC g an 2012–322698) and an NWO Spinoza p ize (NWO SPI 92–266) o C.W.; a Eu opean Union Se en h F amewo k P og amme g an (EU FP7) TANDEM p ojec (HEALTH-F3–2012-305279) o C.W. and V.K.;. A CONACYT-I2T2 schola ship (382117) o R.A-G. The Biobank-Based In eg a i e Omics S udies (BIOS) Conso ium is unded by BBMRI-NL, a esea ch in as uc u e inanced by he Du ch go e nmen (NWO 184.021.007). This wo k was suppo ed by Radboud Uni e si y Medical Cen e Hypa ia Tenu e T ack G an (2018) o Y.L. A ailabili y o da a and ma e ials The decon olu ion summa y s a is ics a e made a ailable as supplemen a y able. In o ma ion on how o eques he geno ype and RNAseq da a used o he eQTL calcula ion can be ound he e: h ps://www.bbm i.nl/acquisi ion-use- analyze/bios. A subse o he single cell eQTLs is p elimina y da a o which a manusc ip is in p epa a ion, and will be made a ailable a e publica ion o ha manusc ip . Con ac Lude F anke ([email p o ec ed]) o eques access o his da a. The GEO accession code o he exp ession da a o 500FG is GSE134080. E hics app o al and consen o pa icipa e We ha e used exis ing and al eady published da a only. The e o e, we did no ge p io e hics app o al o consen o pa icipa e. Consen o publica ion No applicable. Compe ing in e es s The au ho s decla e no compe ing in e es s. Au ho de ails 1 Depa men o Gene ics, Uni e si y o G oningen, Uni e si y Medical Cen e G oningen, G oningen, he Ne he lands. 2 Depa men o Gene ics, Oncode Ins i u e, Uni e si y o G oningen, Uni e si y Medical Cen e G oningen, G oningen, he Ne he lands. 3 Es onian Genome Cen e, Ins i u e o Genomics, Uni e si y o Ta u, Ta u, Es onia. 4 Cen e o Indi idualised In ec ion Medicine (CiiM) & TWINCORE, join en u es be ween he Helmhol z-Cen e o In ec ion Resea ch (HZI) and he Hanno e Medical School (MHH), Feodo -Lynen-S . 7, 30625 Hanno e , Ge many. 5 Uni e si y o G oningen and Uni e si y Medical Cen e G oningen, Genomics Coo dina ion Cen e , G oningen, he Ne he lands. 6 Depa men o Labo a o y Medicine, Labo a o y o Medical Immunology, Radboud Uni e si y Medical Cen e, Nijmegen, he Ne he lands. 7 Depa men o In e nal Medicine and Radboud Cen e o In ec ious Diseases, Radboud Uni e si y Medical Cen e , Nijmegen, he Ne he lands. 8 Depa men o Genomics & Immuno egula ion, Li e and Medical Sciences Ins i u e (LIMES), Uni e si y o Bonn, Bonn, Ge many. Recei ed: 17 Janua y 2020 Accep ed: 1 June 2020 Re e ences 1. Hindo LA, e al. Po en ial e iologic and unc ional implica ions o genome-wide associa ion loci o human diseases and ai s. P oc Na l Acad Sci U S A. 2009;106:9362–7. 2. Ioni a-Laza I, McCallum K, Xu B, Buxbaum JD. A spec al app oach in eg a ing unc ional genomic anno a ions o coding and noncoding a ian s. Na Gene . 2016;48:214–20. 3. Ja ie e BM, e al. Lineage-Speci ic Genome A chi ec u e Links Enhance s and Non-coding Disease Va ian s o Ta ge Gene P omo e s. Cell. 2016;167:1369–1384.e19. 4. Wes a H-J, e al. Sys ema ic iden i ica ion o ans eQTLs as pu a i e d i e s o known disease associa ions. Na Gene . 2013;45:1238–43. Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 22 o 23 5. Joehanes R, e al. In eg a ed genome-wide analysis o exp ession quan i a i e ai loci aids in e p e a ion o genomic associa ion s udies. Genome Biol. 2017;18:16. 6. Raj T, e al. Pola iza ion o he e ec s o au oimmune and neu odegene a i e isk alleles in leukocy es. Science. 2014; 344:519–23. 7. Pe e s JE, e al. Insigh in o Geno ype-Pheno ype Associa ions h ough eQTL Mapping in Mul iple Cell Types in Heal h and Immune-Media ed Disease. PLoS Gene . 2016;12:e1005908. 8. Na anbhai V, e al. Genomic modula o s o gene exp ession in human neu ophils. Na Commun. 2015;6:7545. 9. Chen L, e al. Gene ic D i e s o Epigene ic and T ansc ip ional Va ia ion in Human Immune Cells. Cell. 2016;167:1398– 1414.e24. 10. Wes a H-J, e al. Cell Speci ic eQTL Analysis wi hou So ing Cells. PLoS Gene . 2015;11:e1005223. 11. Zhe nako a DV, e al. Iden i ica ion o con ex -dependen exp ession quan i a i e ai loci in whole blood. Na Gene . 2017;49:139–45. 12. A an D, Hu Z, Bu e AJ. xCell: digi ally po aying he issue cellula he e ogenei y landscape. Genome Biol. 2017;18:220. 13. Ne ea MG, e al. Unde s anding human immune unc ion using he esou ces om he Human Func ional Genomics P ojec . Na Med. 2016;22:831–3. 14. Agui e-Gamboa R, e al. Di e en ial E ec s o En i onmen al and Gene ic Fac o s on T and B Cell Immune T ai s. Cell Rep. 2016;17:2474–87. 15. Bakke OB, e al. In eg a ion o mul i-omics da a and deep pheno yping enables p edic ion o cy okine esponses. Na Immunol. 2018;19:776–86. 16. Tigchelaa EF, e al. Coho p o ile: Li eLines DEEP, a p ospec i e, gene al popula ion coho s udy in he no he n Ne he lands: s udy design and baseline cha ac e is ics. BMJ Open. 2015;5:e006772. 17. Deelen J, e al. Genome-wide associa ion me a-analysis o human longe i y iden i ies a no el locus con e ing su i al beyond 90 yea s o age. Hum Mol Gene . 2014;23:4420–32. 18. Ho man A, e al. The Ro e dam S udy: 2014 objec i es and design upda e. Eu J Epidemiol. 2013;28:889–926. 19. an G ee enb oek MMJ, e al. The c oss-sec ional associa ion be ween insulin esis ance and ci cula ing complemen C3 is pa ly explained by plasma alanine amino ans e ase, independen o cen al obesi y and gene al in lamma ion ( he CODAM s udy). Eu J Clin In es ig. 2011;41:372–9. 20. Schoenmake M, e al. E idence o gene ic en ichmen o excep ional su i al using a amily app oach: he Leiden Longe i y S udy. Eu J Hum Gene . 2006;14:79–84. 21. Willemsen G, e al. The Ne he lands Twin Regis e biobank: a esou ce o gene ic epidemiological s udies. Twin Res Hum Gene . 2010;13:231–45. 22. Bonde MJ, e al. Disease a ian s al e ansc ip ion ac o le els and me hyla ion o hei binding si es. Na Gene . 2017; 49:131–8. 23. Adams D, e al. BLUEPRINT o decode he epigene ic signa u e w i en in blood. Na Bio echnol. 2012;30:224–6. 24. an de Wijs MGP, e al. Single-cell RNA sequencing iden i ies cell ype-speci ic cis-eQTLs and co-exp ession QTLs. Na Gene . 2018;50:493–7. 25. Newman AM, e al. Robus enume a ion o cell subse s om issue exp ession p o iles. Na Me hods. 2015;12:453–7. 26. F iedman J, Has ie T, Tibshi ani R. Regula iza ion pa hs o gene alized linea models ia coo dina e descen . J S a So w. 2010;33:1–22. 27. G uden K, e al. A ‘c ossomics’s udy analysing a iabili y o di e en componen s in pe iphe al blood o heal hy caucasoid indi iduals. PLoS One. 2012;7:e28761. 28. Da enpo EE, e al. Disco e ing in i o cy okine eQTL in e ac ions om a lupus clinical ial; 2017. h ps://doi.o g/10. 1101/118703. 29. Wilson DR, Sun W, Ib ahim JG. Mapping Tumo -Speci ic Exp ession QTLs In Impu e Tumo Samples; 2017. h ps://doi. o g/10.1101/136614. 30. Geelehe P, e al. Cance exp ession quan i a i e ai loci (eQTLs) can be de e mined om he e ogeneous umo gene exp ession da a by modeling a ia ion in umo pu i y. Genome Biol. 2018;19:130. 31. Glas onbu y CA, Al es AC, Mous a a JE-S, Small KS. Cell- ype he e ogenei y in adipose issue is associa ed wi h complex ai s and e eals disease- ele an cell-speci ic eQTLs; 2018. h ps://doi.o g/10.1101/283929. 32. Robinson MD, McCa hy DJ, Smy h GK. edgeR: a bioconduc o package o di e en ial exp ession analysis o digi al gene exp ession da a. Bioin o ma ics. 2010;26:139–40. 33. Dobin A, e al. STAR: ul a as uni e sal RNA-seq aligne . Bioin o ma ics. 2013;29:15–21. 34. Li H, e al. The Sequence Alignmen /Map o ma and SAM ools. Bioin o ma ics. 2009;25:2078–9. 35. Ande s S, Pyl PT, Hube W. HTSeq--a Py hon amewo k o wo k wi h high- h oughpu sequencing da a. Bioin o ma ics. 2015;31:166–9. 36. an Dam RM, Boe JM, Feskens EJ, Seidell JC. Pa en al his o y o diabe es modi ies he associa ion be ween abdominal adiposi y and hype glycemia. Diabe es Ca e. 2001;24:1454–9. 37. Deelen P, e al. Geno ype ha monize : au oma ic s and alignmen and o ma con e sion o geno ype da a in eg a ion. BMC Res No es. 2014;7:901. 38. Howie BN, Donnelly P, Ma chini J. A lexible and accu a e geno ype impu a ion me hod o he nex gene a ion o genome-wide associa ion s udies. PLoS Gene . 2009;5:e1000529. 39. Zhou X, Song Y, Wang L, Liu Q. P econdi ioned GAOR me hods o sol ing weigh ed linea leas squa es p oblems. J Compu Appl Ma h. (2009;224:242–9. 40. Lawson CL, Hanson RJ. Sol ing Leas Squa es P oblem; 1995. Publishe ’sNo e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a ilia ions. Agui e-Gamboa e al. BMC Bioin o ma ics (2020) 21:243 Page 23 o 23