Extracellular vesicle PD‑L1 dynamics predict durable response to immune‑checkpoint inhibitors and survival in patients with non‑small cell lung cancer

deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 h ps://doi.o g/10.1186/s13046‑022‑02379‑1 RESEARCH Ex acellula esicle PD‑L1 dynamics p edic du able esponse oimmune‑checkpoin inhibi o s andsu i al inpa ien s wi hnon‑small cell lung cance Diego de Miguel‑Pe ez1,2, Alessand o Russo2,3, Osca A ie a4, Mu a Ak5,6, Feliciano Ba on4, Mu hukuma Gunaseka an2, P iyada shini Mamindla6, Luis La a‑Mejia4, Ch is ine B. Pe e son7, Mehme E. E 5,6, Vishal Peddagangi eddy5, F ancesco Buemi3, B andon Coope 2, Paolo Manca8, Rena G. Lapidus2, Ru‑Ching Hsia2, And es F. Ca dona9, Aung Naing10, Sunjay Kaushal2, F ed R. Hi sch1, Philip C. Mack1, Ma ia Jose Se ano11, Vincenzo Adamo3, Ri ka R. Colen5,6 and Ch is ian Rol o1,2* Abs ac Backg ound: Immune‑checkpoin inhibi o s (ICIs) changed he he apeu ic landscape o pa ien s wi h lung cance . Howe e , only a subse o hem de i ed clinical bene i and e idenced he need o iden i y eliable p edic i e bio‑ ma ke s. Liquid biopsy is he non‑in asi e and epea able analysis o biological ma e ial in body luids and a p omis‑ ing ool o cance bioma ke s disco e y. In pa icula , he e is g owing e idence ha ex acellula esicles (EVs) play an impo an ole in umo p og ession and in umo ‑immune in e ac ions. Thus, we e alua ed whe he ex acellula esicle PD‑L1 exp ession could be used as a bioma ke o p edic ion o du able ea men esponse and su i al in pa ien s wi h non‑small cell lung cance (NSCLC) unde going ea men wi h ICIs. Me hods: Dynamic changes in EV PD‑L1 we e analyzed in plasma samples collec ed be o e and a 9 ± 1 weeks du ‑ ing ea men in a e ospec i e and a p ospec i e independen coho s o 33 and 39 pa ien s, espec i ely. Resul s: As a esul , an inc ease in EV PD‑L1 was obse ed in non‑ esponde s in compa ison o esponde s and was an independen bioma ke o sho e p og ession‑ ee su i al and o e all su i al. To he con a y, issue PD‑L1 exp ession, he commonly used bioma ke , was no p edic i e nei he o du able esponse no su i al. Conclusion: These indings indica e ha EV PD‑L1 dynamics could be used o s a i y pa ien s wi h ad anced NSCLC who would expe ience du able bene i om ICIs. Keywo ds: Ex acellula esicles, PD‑L1, Bioma ke s, Immuno he apy, NSCLC © The Au ho (s) 2022. 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 eco mmons. o g/ licen ses/ by/4. 0/. The C ea i e Commons Public Domain Dedica ion wai e (h p:// c ea i eco mmons. o g/ publi cdoma in/ 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. Backg ound Immune checkpoin inhibi o s (ICIs) ha e e olu ion- ized he ea men o se e al malignancies, including non-small cell lung cance (NSCLC). P og ammed cell dea h p o ein ligand-1 (PD-L1) exp ession in cance cells is one o he inhibi o y mechanisms in ol ed in umo immune e asion by PD-1 binding and subsequen T cell impai men [1]. Se e al ials ha e epo ed imp essi e Open Access *Co espondence: ch is ian. ol [email p o ec ed] 1 Cen e o Tho acic Oncology, Tisch Cance Ins i u e, Icahn School o Medicine a Moun Sinai, New Yo k, NY, USA Full lis o au ho in o ma ion is a ailable a he end o he a icle Page 2 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 ac i i y o an i-PD-(L)1 monoclonal an ibodies alone o in combina ion wi h chemo he apy o o he immu- no he apeu ic d ugs such as an i-CTLA-4 in pa ien s wi hNSCLC [2–4]. Consequen ly, he FDA app o ed he i s - o second-line use o d ugs such as pemb olizumab (an i-PD-1), a ezolizumab (an i-PD-L1), ni olumab (an i- PD-1), o ecen ly cemiplimab (an i-PD-1) [5] in se e al umo s including lung cance . Signi ican ad ances ha e been made in he sea ch o he ideal p edic i e bioma ke , including he ecen accep ance o umo mu a ional bu den (TMB) and mic osa elli e ins abili y-high (MSI-H) as p edic i e bio- ma ke s in he umo -agnos ic use o pemb olizumab [6]. Howe e , PD-L1 de ec ion by immunohis ochem- is y (IHC) is he FDA-app o ed and mos commonly used p edic i e bioma ke in hese pa ien s. Ne e he- less, many pa ien s exp essing high PD-L1 did no ben- e i om he ea men and a conside able pe cen age o hose wi h low/nega i e PD-L1 exp ession did, which migh be caused by i s high a iabili y [7, 8]. Along wi h in e - umo a iabili y, pa ien s wi hNSCLC show sub- s an ial in a- umo al he e ogenei y and changes in PD-L1 exp ession can occu a e i s -line ea men s, hinde ing he accu a e classi ica ion o PD-L1 s a us [9]. The e o e, aside om he lack o “ eal- ime” in o ma ion, a single issue biopsy may no be able o ecapi ula e he exac s a us o he umo mic oen i onmen a he ime o ea men ha , in some cases, can be mon hs o e en yea s a e umo collec ion. Thus, he e is a huge need o iden i y eliable p edic i e bioma ke s o an i-PD(L)1 agen s ha can e lec he s a us o he umo mic oen i- onmen in eal- ime. PD-L1 p o ein exp ession can also be ound in ex acel- lula esicles (EVs) [10]. These esicles a e double‐mem- b ane s uc u es o 20 – 2000nm in ol ed in in e cellula communica ion and ound in body luids such as blood [11]. EVs a e in ol ed in he c oss- alk wi hin he umo mic oen i onmen and play a ole in he inhibi ion o he an i- umo immune esponse and me as asis, in pa icu- la by PD-L1 p esen a ion [12–15] So a , only ew s udies ha e e alua ed he po en ial o PD-L1 exp ession inEVs as a p edic ing bioma ke in pa ien s wi hlung cance unde going ICIs and hey showed con lic ing esul s ha wa an u he in es iga ion [16, 17]. On he o he hand, adiomics is a apidly g owing ield in imaging, which can con e a pa ien ’s imaging scans in o mineable quan i a i e da a o be e unde - s and he umo he e ogenei y and mic oen i onmen [18]. In pa icula , adiomics may p edic immuno he - apy esponse and ou come in mul iple cance s, includ- ing NSCLC and o he ad anced solid umo s [18–21]. Al hough combining independen p edic i e ma ke s has been ecommended o imp o e accu acy o ea men esponse p edic ion, o he bes o ou knowledge, only one s udy has a emp ed o combine adiomics and liquid biopsy da a o p edic he esponse o immuno he apy in pa ien s wi hNSCLC [16]. Conside ing he limi a ions o ha s udy ega ding a e y small coho , using non-con- as compu ed omog aphy (CT), and mos likely model o e i ing, we belie e ha u he clinical e alua ion o hese combined ma ke s is needed. The e is compelling e idence ha he an i- umo immune esponse is a complex p ocess egula ed by he in e ac ion be ween he umo , he immune sys em, and mul iple hos ac o s in which EVs play a c i ical ole. Thus, we aimed o iden i y and alida e he p edic i e ole o EV PD-L1 dynamics in pa ien s wi had anced/ me as a ic NSCLC ea ed wi h ICIs compa ed o he s anda d-o -ca e issue PD-L1. Addi ionally, we aimed o c ea e and e alua e he pe o mance o a mul ipa ame ic p edic i e model wi h he inclusion o adiomics analysis in ou ini ial coho . Ma e ials andme hods S udy design andpa ien s We conduc ed he e ospec i e analysis o blood sam- ples and CT scan images om pa ien s wi had anced/ me as a ic NSCLC ea ed wi h an i-PD-1 an ibodies a he Medical Oncology Uni o A.O. Papa do o Messina, I aly, be ween May 2018 and No embe 2019 wi h ol- low-up un il Augus 2021 (T aining coho A). Then, we p ospec i ely analyzed blood samples ompa ien s wi h ad anced/me as a ic NSCLC en olled in he phase 2 PROLUNG clinical ial [22], unde going Pemb oli- zumab + Doce axel o Doce axel alone a he Na ional Cance Ins i u e, Mexico, wi h ollow-up un il Augus 2021 (Valida ion coho B) (Fig.1). All pa ien s p o ided w i en in o med consen and he s udy was app o ed by each ins i u ional e iew boa d. Inclusion c i e ia consid- e ed pa ien s olde han 18yea s old a he ime o diag- nosis, s age IIIB o IV acco ding o he 8 h edi ion o he Ame ican Join Commi ee on Cance TNM manual [23], and immune-naï e pa ien s ea ed wi h an i-PD(L)-1 (ni olumab, pemb olizumab) inhibi o s in he i s , sec- ond, o hi d line. His ological PD-L1 exp ession was assessed by immu- nohis ochemis y using SP263, 22C3, o 28–8 clones as pe clinical p ac ice in he mos ecen o malin- ixed pa a in-embedded issue block om he p ima y is- sue o me as asis and classi ied acco ding o he umo p opo ion sco e (TPS). Tumo assessmen du ing ea - men was e alua ed by CT scan a e 9 ± 1weeks and a 21 ± 3weeks o ea men acco ding o he Response E alua ion C i e ia in Solid Tumo s (RECIST) e sion 1.1 [24]. Addi ionally, an independen ins i u ion e iew o he CT scan om he disco e y coho was conduc ed Page 3 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 using RECIST 1.1 and Immune- ela ed Response E alu- a ion C i e ia In Solid Tumo s (i RECIST) [25] o he adiomics analysis. Du able esponde s included hose pa ien s demons a ing comple e esponse (CR), pa ial esponse (PR), o s able disease (SD) a he 21 ± 3 week e alua ion while non- esponde s included hose wi h p og essi e disease (PD). P og ession- ee su i al (PFS) was de ined as he ime elapsed be ween he s a o he ea men o he i s adiological o clinical p og ession and o e all su i al (OS) as he ime om he s a o he he apy ill exi us. Blood samples Th ee mL o pe iphe al blood we e collec ed in o EDTA Vacu aine ® ubes a baseline (T1) and a he i s esponse e alua ion (T2). Blood samples we e cen i- uged a 2,000 × g o 15min and plasma was isola ed and ozen a -80°C. Ex acellula esicle isola ion Ex acellula esicles we e isola ed acco ding o s and- a d p o ocols om ou g oup [26, 27]. B ie ly, plasma aliquo s we e hawed and each500µl we e dilu ed in o 1mL o 1X PBS and cen i uged a 3,000 × g o 20min a 4°C and la e a 10,000 × g o 30min a 4°C o emo e cell deb is. Supe na an s we e eco e ed, dilu ed wi h 1X PBS, and cen i uged a 100,000 × g o 70min a 4°C in o 6.5mL, Open-Top Thickwall Polyca bona e Tubes (Beckman Coul e ) in an Op ima MAX Ul acen i uge (Beckman Coul e ). Finally, EVs pelle s we e esuspended in 1X PBS o lysed wi h 1X RIPA lysis bu e (Cell Signal- ing) and sonica ed o 2min o u he analysis. Ex acellula esicle PD‑L1 cha ac e iza ion Following he las ecommenda ions o he In e na ional Socie y o Ex acellula Vesicles (ISEV) [28], EVs we e cha ac e ized by nanopa icle acking analysis (NTA), ansmission elec on mic oscopy (TEM), and wes e n blo ollowing ou s anda dized me hodology [27]. In addi ion, immunogold TEM cha ac e iza ion o PD-L1 exp ession in EVs was pe o med. EV PD-L1 exp ession was e alua ed by immunoblo and EV PD-L1 dynam- ics (ΔEV PD-L1) we e calcula ed as he exp ession o PD-L1 no malized agains CD9 in he second ime poin di ided by he same alue in he pai ed baseline sample [(PD-L1/CD9) T2 / (PD-L1/CD9) T1]. A ull desc ip ion o he EV cha ac e iza ion me hodology can be ound in he Supplemen a y Me hods. Inc ease in EV PD-L1 was de ined as pa ien s wi h ΔEV PD-L1 > 1 and dec ease in hose wi h ΔEV PD-L1 < 1. Pa ien s wi h lowe olume o quali y o a ailable plasma we e excluded om he s udy (Coho B: 2 pa ien s in he Pemb olizumab + Doce axel and 4 in he Doce axel g oup). Radiomics imaging analysis Radiomics analysis o a ge and non- a ge lesions was execu ed acco ding o ou es ablished me hodology [21]. B ie ly, all lesions we e segmen ed using 3D Slice 4.10.1 module (Slice 4.10.1: Summa y, Highligh s and Changelog—Announcemen s / Release No es—3D Slice Communi y) by a di e en colo label. Addi ional ol- umes o in e es (VOI) o he no mal pec o alis majo muscle we e segmen ed o wi hin-phase no maliza ion. Ten in ensi y-le el his og am ea u es and 195Gy le el co-occu ence ma ix (GLCM) ea u es we e ex ac ed [29, 30]. We calcula ed he a e age, ange, and angula a iance o each ea u e o di e en angles, esul ing in 39 o a ion-in a ian ex u e ea u es calcula ed o each VOI o i e g ay le els. Addi ionally, we compu ed 195 olume-dependen second-o de ea u es by di id- ing each GLCM ea u e by he olume o he segmen ed lesions; he e o e, a o al o 400 adiomics ea u es we e Fig. 1 S udy design: G aphical scheme o pa ien acc ual, ollow‑up, and bioma ke analysis [c ea ed wi h Bio ende .com] Page 4 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 acqui ed. Radiomics analysis was pe o med using ou in-house pipeline in Ma lab ( e sion 2017b; Ma hWo ks Inc) and Phy on P og amming Language ( e sion Phy on 3.7). Bioin o ma ics ands a is ical analysis S a is ical analysis and g aphs we e done using SPSS [SPSS S a is ics o Windows, Ve sion 22.0 (IBM Co p., A monk NY, US), G aphPad P ism Ve sion 8.4 (G aph- Pad So wa e Inc., San Diego CA, US) and R so wa e ( e sion 3.4.0, R Founda ion o S a is ical Compu ing, Vienna, Aus ia). Non-pa ame ic es e alua ed di e - ences be ween a iables. Uni a ia e p edic i e models we e gene a ed wi h logis ic eg ession using glmne unc ion om he glmne package in R so wa e. Reg es- sion analysis was pe o med wi h he leas absolu e sh inkage and selec ion ope a o (LASSO) ea u e selec- ion me hod o ind he mos ele an adiomics ea u es associa ed wi h he esponse [31]. Selec ed ea u es we e en e ed in o eX eme G adien Boos ing (XGBoos ) o build a classi ica ion model o p edic ing o umo esponse o immuno he apy [32]. Finally, lea e-one-ou c oss- alida ion (LOOCV) was applied o assess he obus ness o ou models. Fea u e selec ion, model build- ing, and ecei e ope a ing cha ac e is ics (ROC) analy- ses we e implemen ed using he R packages XGBoos ( e sion 0.6.4.1), ml ( e sion 2.11), and pROC ( e sion 1.9.1). The a ea unde he cu e (AUC) was calcula ed o each ROC cu e and sensi i i y and speci ici y alues we e shown o he op imal cu -poin alue om each cu e, esul ed by selec ing he alue p o iding highe o e all sensi i i y & speci ici y (Youden’s index). Su i al analyses we e pe o med by Kaplan–Meie (log‐ ank es ) and Cox P opo ional‐Haza ds Reg ession wi h backwa d s epwise selec ion o he mul i a ia e model. Two- ailed p alues < 0.05 we e conside ed s a is ically signi ican . Resul s This s udy en olled wo independen coho s o pa ien s wi had anced/me as a ic NSCLC. The aining coho (A) included 33 pa ien s unde going ICIs wi h a median ollow-up o 12.4mon hs ( ange 2.5 – 33.1). The ali- da ion coho (B) en olled 39 pa ien s wi h median ol- low-up o 13.1mon hs ( ange 3.5 – 56.5) om which 24 ecei ed Pemb olizumab + Doce axel and 15 Doce axel alone. Pa ien s cha ac e is ics a e summa ized in Supple- men a y Table S1. EV PD‑L1 cha ac e iza ion Concen a ion, mo phology, size, and speci ic ma k- e s in plasma EVs we e analyzed o p o e hei na u e and pu i y. Fi s , he NTA showed ha plasma EVs had a concen a ion o 2.15 × 108 pa icles/ mL and a mean diame e o 99.4nm (Fig.2A). Second, he immunogold TEM cha ac e iza ion depic ed EVs o simila size wi h posi i e PD-L1 memb anous exp ession (Fig.2B). Thi d, he wes e n-blo e ealed exp ession o PD-L1 and he EV ma ke s CD9 and Flo illin-1 in EVs while absence o he non-EV ma ke GM130, commonly used as con ol Fig. 2 EVs cha ac e iza ion: (A) Nanopa icle acking analysis (NTA) o EVs isola ed om ad anced NSCLC plasma samples showing a concen a ion o 2.15 × 108 pa icles/mL wi h a mode diame e o 68.4 nm. (B) The immunogold ansmission elec on mic oscopy (TEM) depic ed EVs o simila size wi h exp ession o PD‑L1 in he memb ane. (C) Wes e n blo (WB) images e ealed exp ession o PD‑L1, Flo illin‑1, and CD9 in he plasma EVs and lung cance cul u e EVs, while low exp ession o GM130 Page 5 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 o non-EV con amina ion (Fig.2C). Then, PD-L1 and CD9 exp ession we e analyzed in pai ed samples om each pa ien , calcula ing he ΔEV PD-L1 (Supplemen a y Fig. S1). EV PD‑L1 dynamics a e abe e p edic i e bioma ke o du able esponse oICIs han issue PD‑L1 We analyzed he associa ions be ween he EV PD-L1 dynamics and he di e en clinical cha ac e is ics, including age, gende , line o ea men , his ology, is- sue PD-L1 (TPS), e c. in ou h ee subg oups o pa ien s wi h NSCLC, obse ing no s a is ical associa ion wi h any o hem (Supplemen a y Table S2). Then, we exam- ined he po en ial ole o his dynamic bioma ke as a p edic o o du able ICIs esponse compa ed o he s anda d-o -ca e bioma ke , issue PD-L1 (Rep esen a- i e images om esponde s and non- esponde s and he issue and EV PD-L1 cha ac e iza ion a e shown in Fig.3A-C). We ound an inc ease in EV PD-L1 du - ing ea men in non- esponde s in compa ison wi h dec easing le els in esponde s in ou coho A o ICIs pa ien s (p = 0.017) (Fig.3D). Simila ly, in he Pemb oli- zumab + Doce axel g oup, non- esponde s showed a end owa ds inc eased EV PD-L1 in compa ison o esponde s (p = 0.050) (Fig.3E) while no di e ences we e obse ed in he Doce axel ea ed pa ien s (Fig.3F). No di e ences in hese dynamics we e ound be ween he di e en ea men s (Supplemen a y Fig. S2). Mo eo- e , no associa ion was ound be ween he issue PD-L1 exp ession and he du able esponse in any g oup. Thus, ΔEV PD-L1 ou pe o med issue PD-L1 as a p edic i e ac o o iden i ying pa ien s wi h non-du able clinical bene i s om ICIs. ΔEV PD-L1 showed an a ea-unde - he-cu e (AUC) o 77.3% in coho A (Fig. 3G) and 75% in he Pemb olizumab + Doce axel g oup (Fig.3H), while he issue PD-L1 showed only an AUC o 62.7% and 64.1%, espec i ely. Poo p edic i e alues we e obse ed o bo h bioma ke s in he Doce axel g oup (Fig. 3I). When conside ing he ea ly esponse e alu- a ed a he i s CT scan, simila bu no s a is ically sig- ni ican di e ences we e obse ed in Coho A. Howe e , as only one pa ien showed ea ly PD in he Pemb oli- zumab + Doce axel g oup, no signi ican di e ences we e ound (Supplemen a y Fig. S3). Fu he mo e, when he du able esponse was ana- lyzed in all 57 pa ien s unde going ea men wi h ICIs, he dynamics o EV PD-L1 showed di e ences be ween pa ien s wi h PR, SD, and PD (p = 0.009) since i was pos- i i ely co ela ed wi h lesion size (p = 0.040) (Supplemen- a y Fig. S4). Indeed, pa ien s wi h inc eased EV PD-L1 showed an inc ease in lesion size (p = 0.036), bu no asso- cia ion was ound be ween he issue PD-L1 TPS and umo size (p = 0.330) o pa ien s’ esponse (p = 0.561) (Fig. 4A). Mo eo e , inc eased EV PD-L1 iden i ied non- esponde s wi h 73% sensi i i y and 61% speci ic- i y (p = 0.009) (Fig.4B). On he o he hand, high issue PD-L1 was no associa ed wi h a du able esponse ei he when conside ing pa ien s wi h TPS ≥ 50% (p = 0.192) o wi h TPS ≥ 1% (p = 0.370) (Fig.4B). Addi ional sub- g ouped analysis o he p edic i e pe o mance o EV PD-L1 ac oss di e en ypes and lines o he apy o TPS g oups a e shown in Supplemen a y Fig. S5. EV PD‑L1 dynamics a e ap edic i e bioma ke o su i al Du ing he ollow-up o hese coho s, 28 (84.8%) pa ien s p og essed om Coho A, 22 (91.7%) pa ien s unde going Pemb olizumab + Doce axel, and 15 (100%) o hose who unde wen Doce axel in coho B (Sup- plemen a y Table S1). Fi s , we analyzed he p edic i e alue o PFS o he dynamics o EV PD-L1 in coho A. Pa ien s wi h EV PD-L1 dec ease end o expe ience longe PFS han hose wi h inc easing le els (Haza d a io (HR) = 0.36; p = 0.097) (Fig.5A). This was alida ed in pa ien s unde going Pemb olizumab + Doce axel, whe e hose wi h EV PD-L1 dec ease showed longe PFS (HR = 0.18; p = 0.020) (Fig. 5B). To he con a y, no di - e ences in PFS we e obse ed in he Doce axel g oup (HR = 13.3; p = 0.784) (Fig.5C). The mul i a ia e Cox’s eg ession analysis o he o al 57 pa ien s ecei ing ICIs e ealed ha ΔEV PD-L1 was an independen p edic i e bioma ke o PFS, wi h dec eased le els associa ed wi h longe PFS (HR) = 0.45; p = 0.008) while issue PD-L1 exp ession was no (Supplemen a y Table S3) (Fig.5G). Rega ding he mo ali y o hese pa ien s, 19 (57.6%) pa ien s died in coho A. In coho B, 18 (75%) pa ien s died in he Pemb olizumab + Doce axel g oup and 12 (80%) in he Doce axel g oup (Supplemen a y Table S1). As p e iously desc ibed o he PFS, we obse ed ha pa ien s wi h dec easedEV PD-L1 p esen ed longe OS in coho A (HR = 0.22; p = 0.031) (Fig.5D). As a ali- da ion, i was also associa ed wi h longe OS in pa ien s unde going Pemb olizumab + Doce axel (HR = 0.23; p = 0.039) (Fig.5E) and no di e ences we e obse ed in he Doce axel g oup (HR = 1.63; p = 0.202) (Fig.5F). The mul i a ia e Cox’s eg ession analysis o all 57 pa ien s unde going ICIs demons a ed ha hedec ease in EV PD-L1 was an independen p edic i e bioma ke o longe OS (HR = 0.35; p = 0.004); howe e , he issue PD-L1 was no (Supplemen a y Table S4) (Fig.5G). Radiomic ea u es complemen EV PD‑L1 o  hep edic ion o  he esponse Addi ionally, we pe o med an explo a o y analysis o adiomics da a ob ained om baseline CT scans om 27 pa ien s om ou aining coho (A). Among hese pa ien s, 11 (40.7%) we e classi ied as du able esponde s Page 6 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 Fig. 3 EV PD‑L1 dynamics ou pe o med issue PD‑L1 as a p edic o o ICIs esponse: (A) Rep esen a i e axial sec ion compu ed omog aphy (CT) images om a esponde and a non‑ esponde a baseline and du ing ICIs ea men . (B) Examples o immunohis ochemis y mic og aphs o posi i e and nega i e issue PD‑L1 s aining (scale ba s 5 µm) and (C) EV PD‑L1 blo s om a esponde wi h dec easing EV PD‑L1 (0.29) and a non‑ esponde showing an inc ease (1.55). (D) ICIs coho A (n = 33), non‑ esponde s (NR) showed inc eased EV PD‑L1 du ing ea men in compa ison o esponde s (p = 0.017) (Mann–Whi ney U es ). (E) In he alida ion coho , non‑ esponde s unde going Pemb olizumab + Doce axel (n = 24) showed a end owa ds inc eased EV PD‑L1 in compa ison o esponde s (p = 0.050) while hose ea ed wi h Doce axel alone (n = 15) showed no di e ences (p = 0.794) (F) (Mann–Whi ney U es ). (G) As obse ed in he ROC cu e, EV PD‑L1 dynamics was a be e p edic o han issue PD‑L1 TPS wi h an AUC = 74.4% s. 62.6% o he issue (bina y logis ic eg ession). (H) This was also obse ed in he alida ion coho o pa ien s ea ed wi h ICIs wi h AUC = 75% o he EVs s. 64.1% o he issue. (I) In compa ison, simila AUCs we e obse ed in he Doce axel ea ed g oup wi h 54.5% and 59.1%, espec i ely (bina y logis ic eg ession) Page 7 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 by RECIST and 16 (59.3%) by i RECIST, while 15 (55.6%) pa ien s we e conside ed ea ly esponde s by RECIST and 18 (66.7%) by i RECIST, in he i s CT scan. Figu e6A depic s he adiomics pipeline o ea- u e ex ac ion and model selec ion, whe e he mos el- e an ea u es o p edic RECIST du able esponse we e selec ed and combined, esul ing in a model o 6 LASSO ea u es (Supplemen a y Table S5). We compa ed he p edic i e alue o his signa u e wi h he issue and EV PD-L1. We obse ed ha o RECIST esponse, he com- bina ion o ΔEV PD-L1 and he adiomics signa u e was he bes model, able o iden i y non- esponde s wi h an 81.5% accu acy. A he same ime, o du able i RECIST, indi idual ΔEV PD-L1 was he bes p edic i e model wi h 74.1% accu acy (Fig.6B & C). Simila ly, when p edic ing ea ly esponse, he combina ion wi h adiomics also imp o ed he p edic i e accu acy o he dynamics o EV PD-L1 o ea ly RECIST esponse bu no o i RECIST esponse (Supplemen a y Fig. S6). Fu he mo e, hese six ea u es we e used o p edic su i al, which showed ha only low TL_FLV7 was associa ed wi h wo se PFS (HR = 5.52, p = 0.019) (Supplemen a y Fig. S7 & S8). Discussion Ad anced s ages o NSCLC a e cha ac e ized wi h a supp essed immune sys em wi h educed coun s o CD8 + T-cells, esponsible o he immune esponse agains umo s [33]. Consequen ly, ICIs ha e become one o he mos p omising he apeu ic op ions, e o- lu ionizing he he apeu ic landscape o hese pa ien s. Fig. 4 Changes in lesion size o du able esponse co ela ed wi h EV PD‑L1 dynamics in pa ien s unde going ICIs. (A) As obse ed in he co ela ion ma ix, la ge inc eases in he umo lesion we e obse ed in pa ien s wi h inc eased EV PD‑L1 (p = 0.036) (Mann–Whi ney U es ) bu we e independen o he le els o issue PD‑L1 (p = 0.330) (K uskal–Wallis es ). No associa ion was ound be ween he issue PD‑L1 TPS and he umo esponse (p = 0.561) (Chi‑squa e es ). (B) Inc ease in EV PD‑L1 iden i ied non‑ esponde s (p = 0.009), howe e , nei he high issue PD‑L1 TPS > 50% (p = 0.192) o TPS > 1% (p = 0.370) we e associa ed wi h du able esponse (Chi‑squa e es s) Page 8 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 Ne e heless, he e icacy o hese ea men s can s ill be p ima ily imp o ed wi h he adequa e use o eliable p edic i e bioma ke s ha could s a i y which pa ien s would bene i om hem and a oid unnecessa y ad e se e en s o hose who would no de i e bene i . Nowadays, issue PD-L1 is he s anda d-o -ca e o pa ien s a i ica- ion, howe e , i ails a p edic ing he e icacy o ICIs due o se e al echnical and biological issues associa ed wi h Fig. 5 EV PD‑L1 inc ease as a p edic i e bioma ke o PFS and OS. (A) Pa ien s wi h an inc easing EV PD‑L1 (blue) showed a end o sho e PFS (p = 0.097) in he ICIs coho and demons a ed sho e PFS in he Pemb olizumab + Doce axel ea ed g oup (p = 0.020). S ill, no associa ion wi h PFS was obse ed in he Doce axel g oup (p = 0.784) (C). (D) Longe OS was depic ed in pa ien s wi h EV PD‑L1 inc ease (blue) in he ICIs coho (p = 0.031) and he Pemb olizumab + Doce axel g oup (p = 0.038) (E) while no in he Doce axel con ol g oup (p = 0.202) (F) (log‐ ank es s). Numbe o pa ien s a isk o he e en is shown e e y 6 mon hs and he pe cen age o ee o e en (p og ession o dea h) pa ien s is shown a 12 and 24 mon hs. (G) In he 57 pa ien s unde going ICIs, an EV PD‑L1 inc ease was obse ed in hose wi h sho e PFS and OS while issue PD‑L1 was no ( issue PD‑L1 TPS, da k ed = > 50%, ed = 1–49%, pink < 1%, whi e = unknown; a ow = ongoing ea men ; black & whi e squa es ba = OS a e ea men discon inua ion; x = exi us (dea h); o ange ci cles = p og essi e disease; illed da k blue ec angles = EV PD‑L1 inc ease Page 9 o 14 deMiguel‑Pe eze al. J Exp Clin Cance Res (2022) 41:186 PD-L1 IHC [34]. Mo eo e , he high complexi y o he immune landscape o NSCLC sugges ha many ma k- e s migh be in ol ed in he esponse [35] and hence a e needed o i s p edic ion. In his scena io, liquid biopsy holds p omise as he eal- ime cha ac e iza ion o umo s h ough he s udy o molecules ound in human body luids, able o ack lung umo s e olu ion o e ime [11]. Tissue and blood TMB o dynamics o c DNA ha e been p oposed as po en ial bioma ke s. Howe e , hey a e s ill no widely used in clinical p ac ice due o he lack o me hod s anda diza ion and unce ain p edic i e alue. Indeed, ecen analyses o pi o al s udies, includ- ing KEYNOTE-189 [36], KEYNOTE-021 [37], o he ecen esul s o he BFAST coho C s udy [38], aised se e al conce ns on he alidi y o hese be o ehand p omising bioma ke s, equi ing he e alua ion o o he al e na i e ci cula ing ma ke s o e icacy. The e o e, we examined he p edic i e ole o plasma EV PD-L1 exp es- sion in a e ospec i e coho o pa ien s wi had anced/ me as a ic NSCLC unde going ea men wi h ICIs and alida ed i in a p ospec i e analysis o a sub-coho om he phase 2 PROLUNG clinical ial [22]. Fu he mo e, in an explo a o y analysis, we included adiomics da a in a mul ipa ame ic p edic i e model in combina ion wi h he EV PD-L1. Fi s , we demons a ed ha PD-L1 can be exp essed in EVs in plasma om pa ien s wi h ad ancedNSCLC, con- i ming p e ious indings obse ed in o he solid umo s [39]. Then, we analyzed he dynamics o EV PD-L1 le els om baseline o 8weeks o an i-PD-1 ea men in hese pa ien s. We obse ed ha hese dynamic changes we e associa ed wi h du able esponses o he ea men , since non- esponde s showed inc eased le els in compa ison o a dec ease in esponde s, e en co ela ed wi h he change in lesion size. This concu s wi h p e ious s ud- ies on EV PD-L1 in pa ien s and mice models ecei - ing ICIs in which heinc ease in EV PD-L1 co ela ed wi h poo esponse, disease ac i i y, o lesion size [15, 39, 40]. Simila ly, a smalle s udy analyzed he dynam- ics o PD-L1 exp ession bu ocused only on exosomes om 44 pa ien s wi hmelanoma unde going pemb oli- zumab [13]. Con a y o ou esul s, his s udy desc ibed inc eased le els o exosomal PD-L1 in bo h esponde s and non- esponde s by i RECIST, obse ing a highe inc ease du ing he 6 h week o ea men in esponde s bu equal le els a he 9 h week. The idea o inc eased le - els being a p edic o o be e esponse con adic s hei esul s and hose om o he s udies which demons a ed ha exosomal PD-L1 inhibi ed CD8 + T-cell unc ion and acili a ed umo g ow h in in i o and in i o mod- els [13, 15, 41]. This disc epancy could be caused by hei use o he bes esponse by i RECIST as classi ica ion o he clinical esponse, which would include as espond- e s, pa ien s wi h new me as asis o oligop og ession, o e en hose who sho ly esponded and hen p og essed. In addi ion, he di e ence in imepoin s o second blood Fig. 6 Combina ion o adiomics and EV PD‑L1 dynamics o p edic ing du able esponse: (A) Cha ac e is ic pipeline o adiomic analysis including CT scan image segmen a ion, ea u e ex ac ion, and ea u e and model selec ion by machine lea ning. (B) The in oduc ion o he 6‑ ea u es adiomic signa u e in o he ΔEV PD‑L1 p edic i e model o RECIST imp o ed i s pe o mance as obse ed in he conside able inc ease o sensi i i y and speci ici y, wi h an accu acy o 81.5%. (C) On he con a y, he bes model o p edic ion o i RECIST only included he ΔEV PD‑L1 wi h an accu acy o 74.1% (bina y logis ic eg ession)