1
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
www.na u e.com/scien i ic epo s
Cy okine Th esholds in Gingi al
C e icula Fluid wi h Po en ial
Diagnosis o Ch onic Pe iodon i is
Di e en ia ing by Smoking S a us
N. A ias-Bujanda1, A. Reguei a-Iglesias1, M. Alonso-Samped o2, M. M. González-Pe ei o1,
A. Mi a3, C. Balsa-Cas o1 & I. Tomás1
The objec i e o he p esen s udy was o de e mine cy okine h esholds de i ed om p edic i e
models o he diagnosis o ch onic pe iodon i is, di e en ia ing by smoking s a us. Se en y- i e
pe iodon ally heal hy con ols and 75 subjec s a ec ed by ch onic pe iodon i is we e ec ui ed. Six een
media o s we e measu ed in gingi al c e icula luid (GCF) using mul iplexed bead immunoassays.
The models we e ob ained using bina y logis ic eg ession, dis inguishing be ween non-smoke s and
smoke s. The a ea unde he cu e (AUC) and nume ous classi ica ion measu es we e ob ained. Model
cu es we e cons uc ed g aphically and he cy okine h esholds calcula ed o he alues o maximum
accu acy (ACC). The e we e h ee cy okine-based models and h ee cy okine a io-based models, which
p esen ed wi h a bias-co ec ed AUC > 0.91 and > 0.83, espec i ely. These models we e (cy okine
h esholds in pg/ml o he median ACC using boo s apping o smoke s and non-smoke s): IL1alpha
(46099 and 65644); IL1be a (4732 and 5827); IL17A (11.03 and 17.13); IL1alpha/IL2 (4210 and 7118);
IL1be a/IL2 (260 and 628); and IL17A/IL2 (0.810 and 1.919). IL1alpha, IL1be a and IL17A, and hei
a ios wi h IL2, a e excellen diagnos ic bioma ke s in GCF o dis inguishing pe iodon i is pa ien s om
pe iodon ally heal hy indi iduals. Cy okine h esholds in GCF wi h diagnos ic po en ial a e de ined,
showing ha smoke s ha e lowe h eshold alues han non-smoke s.
Pe iodon i is is a public heal h p oblem, as i is highly p e alen and causes disabili y and social inequali y1.
In 2010, se e e pe iodon i is was es ima ed o be he six h mos p e alen disease globally, a ec ing 743 mil-
lion people wo ldwide2. Pe iodon i is is cu en ly being connec ed bidi ec ionally o he pa hogenesis o a ious
condi ions and sys emic diseases o high mo bi-mo ali y such as diabe es, co ona y hea disease, me abolic
synd ome, ch onic espi a o y diseases, heuma oid a h i is, cance , obs e ic complica ions and cogni i e
impai men 3,4.
I is now widely accep ed ha , al hough he ini ia ing ac o is a polymic obial dysbiosis5, he pa hogenesis o
pe iodon i is is d i en by he de elopmen o a ch onic in lamma o y hos immune esponse6,7. The na u e and
ex en o his esponse a e undamen al de e minan s o he suscep ibili y o and p og ession o pe iodon i is6,7.
Cy okines a e soluble p o ein ‘messenge ’ molecules p oduced by a a ie y o cells ha ansmi signals o
o he cells8. Cy okines play a c ucial ole in ini ia ing and sus aining he in lamma o y immune esponse by s im-
ula ing he p oduc ion o seconda y media o s. These media o s, in u n, e oke a cascade o e en s ha ampli y
he in lamma o y esponse and induce he p oduc ion o enzymes ha a e esponsible o he deg ada ion o
connec i e issue and os eoclas ic bone eso p ion9.
Cy okines in e ac and unc ion wi hin a complex and dynamic ne wo k o in e ac ions, a he han being
domina ed by he ac ion o indi idual cy okines8. In ac , an imbalance be ween he p o-in lamma o y and
an i-in lamma o y cy okines de i ed om Th1, Th2, Th17 and T eg lymphocy e subpopula ions is sugges ed as
1O al Sciences Resea ch G oup, Special Needs Uni , Depa men o Su ge y and Medical-Su gical Special ies,
School o Medicine and Den is y, Heal h Resea ch Ins i u e Founda ion o San iago (FIDIS), Uni e sidade de
San iago de Compos ela, Galicia, Spain. 2Depa men o In e nal Medicine and Clinical Epidemiology, Complejo
Hospi ala io Uni e si a io, San iago de Compos ela, Galicia, Spain. 3Cen e o Ad anced Resea ch in Public Heal h,
FISABIO Founda ion, Valencia, Spain. Co espondence and eques s o ma e ials should be add essed o I.T. (email:
[email p o ec ed])
Recei ed: 3 Augus 2018
Accep ed: 7 No embe 2018
Published: xx xx xxxx
OPEN
www.na u e.com/scien i ic epo s/
2
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
being esponsible o pe iodon al b eakdown h ough cellula and humo al hype -immune esponses10. Howe e ,
he educ ionis app oach is p edominan in i o esea ch, as e y ew au ho s ha e analysed he simul aneous
p esence o mo e han 10 cy okines11–13, o mo e han ou cy okine a ios, in gingi al c e icula luid (GCF) om
pe iodon al pa ien s14,15. Acco dingly, mo e e idence is equi ed om mul iple cy okine analyses o inc ease wha
is unde s ood o his complex and dynamic ne wo k8.
On he o he hand, he de ec ion o bioma ke s in GCF o p edic ing he ea ly onse o pe iodon i is o e al-
ua ing he un ea ed o ea ed disease ac i i y is a key challenge in pe iodon ology16–18. The e is, howe e , limi ed
li e a u e on he de elopmen and alida ion o p edic i e models based on GCF cy okine le els o he diagnosis
o p ognosis o pe iodon i is19,20.
Acco dingly, he objec i es o his c oss-sec ional s udy we e:
1. To compa e he le els o 16 cy okines de ec ed in GCF, as well as mul iple cy okine a ios ob ained om
hem, in pe iodon ally heal hy indi iduals and pa ien s wi h ch onic pe iodon i is.
2. To de e mine he diagnos ic alue h esholds de i ed om cy okine-based and cy okine a io-based mod-
els in non-smoke s and smoke s, selec ing hose models wi h a high disc imina o y capaci y o dis inguish
be ween pe iodon al pa ien s and pe iodon ally heal hy con ols.
3. To alida e cy okine-based and cy okine a io-based models in e nally using boo s apping echniques,
desc ibing hei diagnos ic h esholds, as well as appa en and co ec ed measu es o disc imina ion and
classi ica ion.
Ma e ials and Me hods
Selec ion o s udy g oups. A sample o 150 eligible pa icipan s was ec ui ed among 250 consecu i e
pa ien s om he gene al popula ion who we e e e ed o he School o Medicine and Den is y (Uni e sidade de
San iago de Compos ela, Spain) o an e alua ion o hei o al heal h s a us be ween 2013 and 2015. This sample
consis ed o all he cases (pa ien s wi h he a ge condi ion, ha is, 75 subjec s a ec ed by mode a e o se e e
gene alised ch onic pe iodon i is -pe io g oup-) and a andom sample o he noncases (75 pe iodon ally heal hy
con ols -con ol g oup-). Pa ien s we e selec ed i hey ul illed he inclusion c i e ia, which a e de ailed in he
oo no e o Table1.
One p e iously calib a ed, expe ienced den is pe o med all he pe iodon al examina ions. The p obing
pocke dep h (PPD) and clinical a achmen le el (CAL) we e eco ded on all ee h a six si es pe oo h using
a PCP-UNC 15 p obe. Bleeding on p obing (BOP) and bac e ial plaque le el (BPL) da a we e eco ded o he
ull mou h on a bina y scale (p esence/absence) on six si es pe oo h. S anda dised adiog aphs o all ee h we e
ob ained o assess he al eola bone s a us.
The p esence o pe iodon al heal h o mode a e o se e e gene alised ch onic pe iodon i is was es ablished
acco ding o he clinical/ adiog aphic in o ma ion, applying p e iously published c i e ia21,22. Smoking his o ies
we e ob ained using a ques ionnai e, wi h in o ma ion collec ed on smoking s a us (ne e , pas o cu en , he
numbe o mon hs o smoking and he numbe o ciga e es/day). All he answe s we e e iewed wi h he subjec
by a membe o he s udy eam.
This s udy was conduc ed acco ding o he p inciples ou lined in he Decla a ion o Helsinki (as e ised in
2000) on expe imen a ion in ol ing human beings23. The TRIPOD guidelines we e conside ed o u he p e-
dic i e analysis24.
Gingi al c e icula luid sampling. The GCF collec ion ook place one week a e he ini ial examina ion,
and he samples we e ob ained a he same ime o day (in he a e noon, app oxima ely 5–7 h a e oo hb ush-
ing). A pape s ip (Pe iopape , Ami y ille, NY, USA) was inse ed in o he gingi al sulcus o pe iodon al pocke
o 30 sec, using a GCF collec ion p o ocol p e iously desc ibed25. GCF samples om he con ols and pe iodon-
al pa ien s we e collec ed and pooled om 20 non-adjacen p oximal si es. In he i s case, samples we e aken
om subgingi al si es om ee h in quad an s 1 and 3, and in he second case om si es om he mos in-dep h
PPD in each quad an .
S ips om each subjec we e inse ed in o labelled ubes wi h 300 ml o 0.01 M PBS (pH = 7.2) and a p o ease
inhibi o (Comple e Mini, p o ease inhibi o cock ail able s, Roche Applied Science, Indianapolis, IN, USA).
To ensu e sample collec ion, he GCF olume was de e mined based on measu emen s o weighing he ubes
and s ips be o e and a e sampling using a e y sensi i e scale26 ( eadabili y o 0.01 mg; Explo e Semi Mic o
Ex125M, OHAUS, G ei ensee, Swi ze land). All he samples collec ed had olumes o GCF ≥ 10 µl. A e ob ain-
ing he supe na an , he GCF samples we e ozen a −80 °C un il u he biochemical analysis.
Quan i ica ion o cy okines in gingi al c e icula luid using mul iplexed bead immuno-
assays. GCF cy okine le els we e de e mined using he human cy okine 16-plex P oca a immunoassay
(A yme ix, Inc., San a Cla a, CA, USA). Six een media o s we e measu ed: g anulocy e-mac ophage colony
s imula ing ac o – GMCSF; IFNgamma; IL1alpha; IL1be a; IL2; IL3; IL4; IL5; IL6; IL10; IL12p40; IL12p70; IL13;
IL17A; IL17F; and TNFalpha.
A single in es iga o blinded o he clinical da a pe o med he expe imen al analyses o he GCF cy okine
quan i ica ion. The assays we e pe o med in 96-well il e pla es ollowing he manu ac u e ’s ins uc ions and
applying an analysis p o ocol desc ibed p e iously25. The GCF samples we e quan i ied using he Luminex 100™
ins umen (Luminex Co po a ion, Aus in, Texas, USA) and all o hem we e un in duplica e. The concen a-
ions o he unknown samples we e es ima ed om he s anda d cu e using a 5PL algo i hm and he Luminex IS
2.3 and xPONENT 3.1 so wa e packages (Luminex So wa e, Inc.). Values we e exp essed as pg/ml adjus ing o
www.na u e.com/scien i ic epo s/
3
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
he dilu ion ac o . Samples below he de ec ion limi (DL) o he assay we e eco ded as DL/227, while hose abo e
he uppe limi o quan i ica ion o he s anda d cu es we e assigned he highes alue o he cu e.
S a is ical analysis. The s a is ical analyses we e pe o med using he R so wa e ( e sion 3.4.3)28. I is a ail-
able as F ee So wa e unde he e ms o he F ee So wa e Founda ion’s GNU Gene al Public License in sou ce
code o m. A e applying he Shapi o-Wilks es and e i ying he non-no mal dis ibu ion o almos all he
clinical a iables, he Mann-Whi ney U es was used o compa e he quan i a i e a iables be ween he pe io and
con ol g oups. The Fishe ’s exac es was used o assess he associa ion o he quali a i e a iables be ween bo h
s udy g oups. The signi icance le el applied was a p- alue < 0.05.
Compa ison o GCF cy okine le els and cy okine a io alues in pe iodon ally heal hy indi id-
uals and pa ien s wi h ch onic pe iodon i is. A e e i ying he non-no mal dis ibu ion o a iables
using he Shapi o-Wilks es , he Mann-Whi ney U es was used o compa e he cy okine le els and cy okine
a ios in he con ol and pe io g oups. The signi icance le els applied we e adjus ed by he Benjamini-Hochbe g
co ec ion29, wi h p- alues ≤ 1 × 10−3 and <1 × 10−5, espec i ely. A o al o 66 cy okine a ios we e e alua ed,
aking in o accoun exclusi ely hose cy okines ha showed signi ican le els in he pe iodon al pa ien s com-
pa ed o he con ols (Fig.1).
P edic i e modelling o ch onic pe iodon i is based on cy okine le els and cy okine a ios:
model selec ion; disc imina ion and classi ica ion measu es; de e mina ion o diagnos ic
h esholds; and in e nal alida ion. To ob ain speci ic diagnos ic h esholds di e en ia ing by smok-
ing s a us, we decided o de elop di e en models o non-smoke s and smoke s (n = 93 and 54, espec i ely).
Models we e cons uc ed by selec ing one cy okine o cy okine a io as a p edic o a iable, which was ea ed in
i s o iginal scale.
Clinical Pa ame e s
S udy G oupsa
Con ol g oup
(n = 74) Pe io g oup
(n = 73) P Value
Age (yea s) 45.65 (12.37) 51.12 (10.01) 0.005
Gende
Male 31 31 NS
Female 43 42
No. o ee h 26.72 (3.25) 25.55 (4.00) NS
Full mou h
BPL (%) 26.41 (18.66) 53.08 (26.77) <0.001
BOP (%) 15.05 (6.61) 51.12 (20.07) <0.001
PPD (mm) 2.11 (0.27) 3.49 (0.65) <0.001
CAL (mm) 2.36 (0.46) 4.25 (1.12) <0.001
Sampled si es
BOP (%) 10.11 (10.24) 66.97 (23.93) <0.001
PPD (mm) 2.23 (0.22) 5.65 (0.89) <0.001
CAL (mm) 2.31 (0.27) 6.05 (1.12) <0.001
Smoking habi b
Non-smoke s 61 32 0.001
Smoke s 13 41
Ciga e es/day (no.) 8.08 (4.44) 15.20 (7.94) 0.001
Mon hs o smoking (no.) 236.38 (155.91) 320.78 (109.03) NS
Table 1. Age, gende , smoking habi and clinical cha ac e is ics associa ed wi h he pe iodon al s a us in he
con ol and pe io g oups. Values a e means (s anda d de ia ions) and numbe o subjec s. BPL = bac e ial
plaque le el; BOP = bleeding on p obing; PPD = p obing pocke dep h; CAL = clinical a achmen le el;
NS = no signi ican . In his se ies, he inclusion c i e ia applied we e: 1) age 30 o 75; 2) no medical his o y o
diabe es melli us o hepa ic o enal disease, o o he se ious medical condi ions o ansmi able diseases; 3)
no his o y o alcohol o d ug abuse; 4) no p egnancy o b eas eeding; 5) no in ake o sys emic an imic obials
du ing he p e ious six mon hs; 6) no in ake o an i-in lamma o y medica ion in he p e ious ou mon hs;
7) no ou ine use o o al an isep ics; 8) no p esence o implan s o o hodon ic appliances; 9) no p e ious
pe iodon al ea men ; 10) smoke s who had s opped smoking less han i e yea s be o e he sampling; and
11) he p esence o a leas 18 na u al ee h. aO he 150 pa ien s who ul illed he inclusion c i e ia and had
an adequa e pe iodon al diagnosis, h ee we e excluded o unexpec ed e en s. The con ol g oup included
pe iodon ally heal hy indi iduals wi h BOP < 25%, no si es wi h a PPD ≥ 4 mm and no adiog aphic e idence
o al eola bone loss. Conside ing p e iously es ablished c i e ia21,22, pa ien s in he pe io g oup we e diagnosed
wi h mode a e o se e e gene alised ch onic pe iodon i is. bA pa ien was de ined as a smoke i he/she was
cu en ly smoking and had been a smoke o a leas eigh yea s) and as a non-smoke i he/she had ne e
smoked o had s opped smoking mo e han i e yea s be o e he sampling.
www.na u e.com/scien i ic epo s/
4
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
The s a is ical c i e ion applied o he model selec ion was he abili y o each cy okine- o cy okine a io-based
model o de e mine he p esence o ch onic pe iodon i is using he alue o he a ea unde he cu e (AUC)30.
The AUC alues and hei co esponding 95% con idence in e als (CIs) ob ained by boo s apping we e calcu-
la ed using he pROC package ( e sion 1.10.0)31. Only hose models ha p esen ed an appa en AUC ≥ 0.85 in
bo h ypes o model o smoke s and non-smoke s we e selec ed32.
The bes cu -o alue o op imal classi ica ion h eshold o each model was de ined as ha which p o ides
he maximum pe cen age o co ec p edic ions (accu acy, ACC), and was calcula ed using he P esenceAbsence
package ( e sion 1.1.9)33. By se ing his op imal alue, a ious classi ica ion measu es such as he ACC, sensi i -
i y, speci ici y, posi i e p edic i e alue (PPV), and nega i e p edic i e alue (NPV), as well as hei co espond-
ing 95% CIs acqui ed by boo s apping, we e ob ained using he pROC package31. The espec i e cy okine le els
o cy okine a ios we e calcula ed o all he pe iodon i is p obabili y alues o each model, and he model cu es
we e cons uc ed g aphically using he ggplo package ( e sion 2.2.1)34.
Rega ding in e nal alida ion, boo s apping was used o es o possible o e i ing by de e mining he op i-
mism alues on he disc imina ion and classi ica ion measu es. The boo s ap analysis was eplica ed on 10,000
andom samples o he same sample size, d awn wi h eplacemen s om he o iginal sample35,36. Bias-co ec ed
(bc) AUC and all o he classi ica ion measu es (bc-sensi i i y, bc-speci ici y, bc-PPV, bc-NPV) we e calcula ed as
hei co esponding appa en measu es de i ed om he en i e o iginal sample minus op imism35,36. This ech-
nique was also used o de ine he cy okine h esholds o he median ACC alues de i ed om 10,000 samples
om each model selec ed, as well as he h esholds o he 90% CIs o he ACC alues (Fig.1).
E hics app o al and consen o pa icipa e. The s udy’s p o ocol was app o ed by he Clinical Resea ch
E hics Commi ee o Galicia (numbe 2015/006). Pa ien s who ag eed o pa icipa e in he esea ch p o ided
w i en in o med consen .
Resul s
The mean age o he s udy g oup was 48.37 ± 11.55 yea s; 62 indi iduals we e male and 85 emale. The pe io
g oup had signi ican ly highe BPL, BOP, PPD and CAL alues han he con ol g oup a bo h he ull mou h and
sampling si e le els (p < 0.001; Table1). The numbe o smoke s was signi ican ly highe in he pe io g oup han
in he con ol g oup (41 and 13 pa ien s, espec i ely, p < 0.001; Table1).
Compa ison o GCF cy okine le els and cy okine a io alues in pe iodon ally heal hy indi id-
uals and pa ien s wi h ch onic pe iodon i is. All he p o-in lamma o y cy okines analysed (GMCSF,
IL1alpha, IL1be a, IL6, IL12p40, IL17A, IL17F and TNFalpha), as well as ou cy okines wi h an i-in lamma o y
Figu e 1. Flow cha o he s a is ical analysis: bina y logis ic eg ession and diagnos ic h esholds. AUC:
a ea unde he cu e; ACC: accu acy; Sens: sensi i i y; Spec: speci ici y; PPV: posi i e p edic i e alue; NPV:
nega i e p edic i e alue; CIs: con idence in e als.
www.na u e.com/scien i ic epo s/
5
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
e ec s (IFNgamma, IL2, IL3 and IL4), had signi ican ly highe le els in he pe io g oup han in he con ol
g oup (adjus ed p- alue ≤ 1 × 10−3). Nine een cy okine a ios showed signi ican di e ences be ween he con-
ol and pe io g oups (adjus ed p- alue < 1 × 10−5). O hese a ios, nine we e based on he combina ion o wo
p o-in lamma o y cy okines, which we e: IL1alpha combined wi h GMCSF, IL12p40 and TNFalpha; and IL1be a
combined wi h GMCSF, IL12p40, IL17F o TNFalpha, GMCSF/IL17A and IL17A/IL17F. The emaining 10 a ios
we e based on he combina ion o one p o-in lamma o y cy okine and one cy okine wi h an i-in lamma o y
e ec s. These we e: IL1alpha/IL2; IL1alpha/IL3; IL1alpha/IFNgamma; ILbe a/IL2; ILbe a/IL3; ILbe a/IL4; ILbe a/
IFNgamma; IL17A/IL2; IL17A/IL3; and IL17A/IFNgamma. All hese cy okine a ios, excep o GMCSF/IL17A,
had signi ican ly highe alues in he pe io g oup han in he con ol g oup (Table2).
P edic i e modelling o ch onic pe iodon i is based on GCF cy okine le els and cy okine
a ios: model selec ion; disc imina ion and classi ica ion measu es; de e mina ion o diagnos-
ic h esholds; and in e nal alida ion. The e we e h ee cy okine-based models and h ee cy okine
a io-based models, which had an appa en AUC ≥ 0.85 o bo h non-smoke s and smoke s. These models we e
IL1alpha, IL1be a, IL17A, IL1alpha/IL2, IL1be a/IL2 and IL17A/IL2.
Appa en and bc-pe cen ages o disc imina ion and classi ica ion o he six p edic i e models a e desc ibed
in Table3. The cy okine-based models had AUC and bc-AUC alues ≥ 0.940 and ≥ 0.912, espec i ely, and he
cy okine a io-based model alues we e ≥ 0.857 and ≥ 0.834, espec i ely. The bc-ACC ange de i ed om he
CGF Cy okine
Concen a ion (pg/ml)
Median (IQR)
Con ol g oup Pe io g oup Adjus ed
p- alue
GMCSF 150.24 (129.67) 247.24 (255.55) 7.29E-05
IL1alpha 30405.78 (20713.06) 148825.83 (221175.10) 2.64E-21
IL1be a 2881.75 (1958.43) 17947.50 (17308.08) 5.59E-21
IL6 166.71 (163.90) 313.45 (461.34) 3.77E-06
IL12p40 7.34 (5.65) 17.30 (11.06) 7.10E-12
IL17A 7.53 (9.38) 28.45 (25.83) 9.70E-19
IL17F 3.40 (7.01) 9.62 (15.47) 7.75E-09
TNFalpha 6.46 (13.49) 22.75 (15.95) 2.50E-04
IFNgamma 4.71 (3.85) 9.55 (9.60) 3.42E-07
IL2 9.96 (8.74) 14.96 (8.08) 0.001
IL3 54.26 (37.15) 99.60 (89.75) 1.23E-05
IL4 5.96 (21.42) 12.37 (47.02) 2.50E-04
CGF Cy okine Ra io
GMCSF/IL17A 26.57 (25.02) 7.76 (14.82) 2.80E-08
IL1alpha/GMCSF 218.44 (198.78) 631.24 (1434.02) 1.03E-11
IL1alpha/IL12p40 4531.12 (3831.16) 6849.23 (17879.96) 2.78E-06
IL1alpha/TNFalpha 4524.95 (5930.96) 8427.24 (22799.21) 7.71E-06
IL1be a/GMCSF 19.52 (24.42) 67.75 (97.45) 1.18E-12
IL1be a/IL12p40 402.43 (516.02) 844.61 (1378.24) 2.83E-08
IL1be a/IL17F 662.10 (964.59) 1446.37 (3287.34) 1.62E-06
IL1be a/TNFalpha 468.80 (583.89) 854.64 (1669.85) 2.49E-07
IL17A/IL17F 0.88 (2.46) 2.31 (4.79) 1.88E-06
IL1alpha/IL2 3279.43 (3601.17) 8890.94 (17774.01) 1.93E-14
IL1alpha/IL3 630.99 (759.95) 2262.85 (5590.86) 1.42E-10
IL1alpha/IFNgamma 7728.27 (4744.27) 18426.38 (52652.97) 3.05E-12
IL1be a/IL2 260.08 (210.23) 1249.52 (1480.05) 4.16E-14
IL1be a/IL3 66.56 (87.48) 206.71 (469.16) 8.04E-11
IL1be a/IL4 536.43 (702.25) 1530.57 (2756.95) 7.25E-06
IL1be a/IFNgamma 805.90 (387.53) 2084.77 (5776.57) 5.54E-13
IL17A/IL2 0.64 (0.58) 1.96 (1.41) 5.54E-13
IL17A/IL3 0.13 (0.12) 0.34 (0.32) 4.80E-11
IL17A/IFNgamma 1.81 (1.03) 3.14 (2.03) 4.20E-12
Table 2. Concen a ions o cy okines and cy okine a ios ha showed signi ican di e ences (adjus ed p-
alues ≤ 1 × 10−3 and <1 × 10−5, espec i ely) be ween he con ol and pe io g oups. IQR, in e qua ile ange;
CGF, c e icula gingi al luid. Concen a ion ange o each bioma ke analysed: GMCSF, 0.53–55,050 pg/ml;
IFNgamma, 0.02–6,650 pg/ml; IL1alpha, 0.34–28,800 pg/ml; IL1be a, 0.09–23,150 pg/ml; IL2, 0.04–13,700 pg/
ml; IL3, 0.19–26,500 pg/ml; IL4, 0.10–29,250 pg/ml; IL5, 0.04–17,800 pg/ml; IL6, 0.10–27,200 pg/ml; IL10,
0.04–10,050 pg/ml; IL12p40, 0.14–27,350 pg/ml; IL12p70, 0.26–18,050 pg/ml; IL13, 0.34–23,700 pg/ml; IL17A,
0.36–30,900 pg/ml; IL17F, 0.25–34,700 pg/ml; TNFalpha, 0.21–16,800 pg/ml.
www.na u e.com/scien i ic epo s/
6
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
cy okine-based models was 86.8–94.1% and ha o he cy okine a io-based models was 72.9–88.7%, wi h IL17A
and IL17A/IL2 being he bioma ke s wi h he lowes bc-ACC alues in bo h smoke s and non-smoke s. The 95%
CIs o he model coe icien s and hose o he pe o mance measu es a e de ailed in Supplemen a y Da ase 1.
The pe iodon i is p obabili y ange o he median ACC alues a ied be ween 23 and 51%. The cy okine
h esholds in pg/ml o he median ACC alues (and hose o he 95% CIs o he ACC alues) o smoke s
and non-smoke s we e, espec i ely: IL1alpha model: 46099 (37495–64161) and 65644 (51310–76700); IL1be a
model: 4732 (3705–6459) and 5827 (4721–7532); IL17A model: 11.03 (7.28–15.22) and 17.13 (13.10–22.53);
IL1alpha/IL2 model: 4210 (3164–5648) and 7118 (4798–10166); IL1be a/IL2 model: 260 (63–487) and 628 (348–
897); and IL17A/IL2 model: 0.810 (0.707–1.132) and 1.919 (1.073–3.489). The ange o cy okine h esholds ep-
esen ed a ound 9–13% o he cy okine o a io measu emen ange, excep o IL17/IL2 o non-smoke s (30%).
Compa ed o he non-smoke s, he smoke s had lowe diagnos ic h esholds on all he p edic i e models o bo h
appa en ACC alues and ACC alues ob ained by boo s apping (Figs2–4).
Discussion
High cy okine concen a ions and cy okine a ios in he gingi al c e icula luid o pa ien s
wi h ch onic pe iodon i is. As men ioned in he In oduc ion, he e has been a ailu e o s udy a b oade
spec um o cy okines ha may di ec ly in luence he local in lamma o y esponse in di e en ypes o pe io-
don i is37. The p esen se ies is he i s compa a i e analysis o mo e han 50 cy okine a ios de i ed om he
simul aneous quan i ica ion o 16 cy okines wi h di e en oles in he pa hophysiology o ch onic pe iodon i is7,8.
I should be no ed ha a pa icula ly s ic co ec ed signi icance alue was applied (adjus ed p- alue < 1 × 10−5)
in o de o selec he cy okine a ios wi h he mos signi ican impac on ch onic pe iodon i is. This s a is ical
decision condi ioned he a ios conside ed o be non-signi ican and signi ican . As a consequence, compa isons
wi h he con ibu ions o o he au ho s mus be in e p e ed wi h cau ion.
Al hough e y ew au ho s ha e in es iga ed he a ios be ween p o-in lamma o y cy okines in pe iodon-
i is38–40, up o eigh p o-in lamma o y cy okine a ios showed signi ican di e ences in pe iodon al pa ien s.
Al hough we de ec ed signi ican ly ele a ed le els o all he p o-in lamma o y cy okines analysed, IL1alpha and
IL1be a we e he mos impo an bioma ke s in e ms o inc eased concen a ion associa ed wi h he disease.
This esul ed ha he a ios based on IL1alpha combined wi h GMCSF, IL12p40 o TNFalpha, and IL1be a com-
bined wi h GMCSF, IL12p40, IL17F o TNFalpha, showed signi ican ly highe alues in he pe iodon al pa ien s.
Coinciding wi h he esul s epo ed by Azman e al.41, we also ob ained a signi ican ly ele a ed IL17A/IL17F
a io in he pa ien s wi h ch onic pe iodon i is. In e es ingly, in his se ies, and unlike he o he p o-in lamma o y
cy okine a ios, he GMCSF/IL17A a io had signi ican ly lowe alues in he pe iodon al pa ien s, ep esen ing
he i s e idence o he impac o his a io in he pa hogenesis o pe iodon i is.
Mos p e ious s udies ha e ocused on he analysis o a ios be ween p o-in lamma o y and an i-in lamma o y
cy okines, o ice e sa, in pe iodon al diseases, wi h IL1be a/IL10 and IL11/IL17 being he mos
Model Smoking
s a us AUC ACC
(%) Sensi i i y
(%) Speci ici y
(%) Posi i e P edic i e
Value (%) Nega i e P edic i e
Value (%)
IL1alpha
Smoke 0.966
0.951 92.5
89.4 100.0
97.1 69.2
65.8 91.1
89.9 100.0
92.6
Non-smoke 0.959
0.958 93.5
92.4 87.5
85.8 96.7
96.0 93.5
92.1 93.7
92.8
IL1be a
Smoke 0.968
0.945 94.4
90.7 97.5
96.6 84.6
71.1 95.2
91.9 92.3
88.8
Non-smoke 0.944
0.943 94.6
94.1 90.6
89.5 96.7
96.5 93.7
93.4 95.2
94.6
IL17A
Smoke 0.940
0.912 92.5
90.3 95.1
93.9 84.6
79.1 95.2
93.6 85.7
82.1
Non-smoke 0.914
0.914 88.1
86.8 78.1
76.1 93.4
92.5 86.2
84.3 89.2
88.2
IL1alpha/IL2
Smoke 0.878
0.868 85.1
81.1 100.0
99.0 38.4
29.5 83.6
80.2 100.0
99.0
Non-smoke 0.911
0.905 88.1
85.1 84.3
80.0 90.1
88.6 81.8
78.6 91.6
89.2
IL1be a/IL2
Smoke 0.906
0.896 92.5
88.7 95.1
94.7 84.6
76.3 95.1
91.1 85.7
82.1
Non-smoke 0.886
0.881 84.9
79.5 81.2
70.6 86.8
84.3 76.4
72.3 89.8
84.1
IL17A/L2
Smoke 0.955
0.948 92.5
87.2 100.0
98.6 69.2
51.4 91.1
86.5 100.0
95.7
Non-smoke 0.857
0.834 80.6
72.9 87.5
81.7 77.0
68.3 66.6
54.3 92.3
89.1
Table 3. Appa en and bias-co ec ed measu es o disc imina ion and classi ica ion o he models based on
cy okines and ci okine a ios o bo h smoke s and non-smoke s. In each cell, he i s alue is e e ed o
he appa en pe o mance measu es and he second o he co ec ed pe o mance measu es by he le el o
op imism, calcula ed using a boo s ap p ocedu e. The 95% CIs o he di e en pe o mances measu es a e
de ailed in Supplemen a y Da ase 1.
www.na u e.com/scien i ic epo s/
7
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
e alua ed14,38,42–45. In he p esen se ies, up o nine a ios based on he combina ion o one p o-in lamma o y
cy okine (IL1alpha, ILbe a o IL17A) and one cy okine wi h an i-in lamma o y e ec s (IFNgamma, IL2, IL3 o
IL4) showed signi ican ly highe alues in he pe iodon al pa ien s. These esul s we e due o he highe mean
concen a ions o p o-in lamma o y cy okines compa ed o he le els p esen ed by an i-in lamma o y cy okines.
In con as o he indings o S adle e al.46, hese media o s also showed a signi ican mean inc ease associa ed
wi h ch onic pe iodon i is. On he o he hand, applying mul i a ia e p edic i e modelling echniques, we ha e
p e iously demons a ed ha he ex en o he pe iodon i is-associa ed imbalance be ween IL1alpha, ILbe a o
IL17A (ac ing as enhance s) and IFNgamma, IL2, IL3 o IL4 (ac ing as p o ec o s) was associa ed wi h a pa icu-
la p obabili y o ha ing ch onic pe iodon i is25.
We ha e no ound any a icles ha would enable us o compa e ou indings on a ios be ween IL1alpha
and di e en an i-in lamma o y cy okines. Rega ding he a ios be ween ILbe a and o he an i-in lamma o y
cy okines, some au ho s ha e obse ed ha : he ILbe a/IL10 a io was inc eased in he GCF o gingi al issue
o pa ien s wi h agg essi e pe iodon i is o ch onic pe iodon i is14,42,43; his a io was signi ican ly educed a e
pe iodon al he apy43. Howe e , a e s udying hese pape s in de ail, hese esul s can be a ibu ed mainly o
signi ican ly highe mean le els o IL1be a, while he le els o IL10 showed non-signi ican indi idual a ia ions.
These esul s ob ained in i o call in o ques ion he impo ance o his a io in he pa hogenesis o pe iodon i is.
Likewise, in he p esen s udy, no signi ican di e ences in IL10 le els be ween he con ols and pe iodon al
Figu e 2. Model cu es based on IL1alpha and IL1alpha/IL2, de ining he diagnos ic h esholds o appa en
and median ACC alues, as well as hose h esholds o he 90% CIs o he ACC alues.
www.na u e.com/scien i ic epo s/
8
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
pa ien s we e de ec ed, and so he IL1be a/IL10 a io was no e alua ed. Howe e , i should be no ed ha IL10
acqui ed a g ea e p o agonism as an an i-in lamma o y cy okine wi hin a wo-bioma ke p edic i e model, as
his inc eased he capaci y o IL1be a o disc imina e he ch onic pe iodon i is s a e25. In con as , in he p esen
se ies, we obse ed ha o he a ios, such as IL1be a/IFNgamma, ILbe a/IL2, ILbe a/IL3 and ILbe a/IL4, may
play an essen ial ole in quan i a i e e ms in ch onic pe iodon i is. Se e al s udies ha e e ealed ha he IL11/
IL17 a io was educed in pa ien s wi h ch onic and agg essi e pe iodon i is44,45,47, al hough o he au ho s ha e
desc ibed con lic ing indings38. In he p esen s udy, o he a ios such as IL17A/IFNgamma, IL17A/IL2, IL17A/
IL3 and IL17A/IL4 had signi ican ly highe alues in he pe iodon al pa ien s, e lec ing hei impac on ch onic
pe iodon i is.
Consequen ly, his s udy is he i s ime ha e idence is p o ided on a high numbe o a ios be ween
p o-in lamma o y cy okines o p o-in lamma o y and an i-in lamma o y cy okines ha , due o hei pe o -
mance in GCF samples, could be bioma ke s associa ed wi h ch onic pe iodon i is. Fu u e esea ch is equi ed o
cla i y he ele ance o hese a ios in he ch onic pe iodon i is pa hogenesis.
High p edic i e abili y o GCF cy okine le els and cy okine a ios o he diagnosis o ch onic
pe iodon i is. Due o he cha ac e is ics o cy okine ne wo ks48, whe he cy okines in GCF may show an
accep able abili y o disc imina e ch onic pe iodon i is om pe iodon al heal h is ques ioned. Howe e , his
Figu e 3. Model cu es based on IL1be a and IL1be a/IL2, de ining he diagnos ic h esholds o appa en and
median ACC alues, as well as hose h esholds o he 90% CIs o he ACC alues.
www.na u e.com/scien i ic epo s/
9
SCIENTIFIC REPORTs | (2018) 8:18003 | DOI:10.1038/s41598-018-35920-4
a i ma ion is suppo ed by li le e idence, as he e a e e y ew s udies ha ha e e alua ed he p edic i e p op-
e ies o cy okines in ch onic and agg essi e pe iodon i is using an app op ia e expe imen al design19,20. The
cu en se ies e eals he i s esul s on he p edic i e abili y o cy okines and cy okine a ios o he diagnosis
o ch onic pe iodon i is, di e en ia ing be ween smoke s and non-smoke s. Mo eo e , in e nal alida ion was
ca ied ou o he i s ime on he p edic i e pa ame e s ob ained, as ecommended in he TRIPOD guidelines24.
In his s udy, in ela ion o indi idual cy okines, and co obo a ing obse a ions published p e iously by
ou esea ch g oup25, he e we e h ee models consis ing o IL1alpha, IL1be a and IL17A, which p esen ed a
bc-AUC > 0.90 o bo h smoke s and non-smoke s. Acco ding o expe s in he ield32, hese high AUC al-
ues indica e ha hese p o-in lamma o y cy okines ha e a g ea capaci y o disc imina e he disease condi ion.
Consequen ly, hese p o-in lamma o y cy okines we e associa ed wi h ele a ed bc-ACC pe cen ages: 90.7% ( o
IL1be a), 90.3% ( o IL17A) and 89.4% ( o IL1alpha) in smoke s; and 94.1%, 86.8% and 92.4%, espec i ely, in
non-smoke s. Findings on IL1’s high p edic i e abili y a e consis en wi h hose p e iously desc ibed by Baeza e al.20,
while IL17’s indings ep esen he i s e idence o a s ong diagnos ic capabili y. In ou opinion, ou esul s on
he high p edic i e po en ial o hese cy okines a e compa able o hose ound o o he well-known bioma ke s,
such as di e en me allop o einases20.
We e alua ed he cy okine a ios using p edic i e modelling echniques, wi h he aim being o iden i y a se o
bioma ke s ha gua an ee a high diagnos ic p edic abili y16. In his sense, we ob ained h ee a io-based models
Figu e 4. Model cu es based on IL17A and IL17A/IL2, de ining he diagnos ic h esholds o appa en and
median ACC alues, as well as hose h esholds o 90% CIs o he ACC alues.
