identification of novel biomarkers for predicting

Research ArticleIdentification of Novel Biomarkers for Predicting Prognosis andImmunotherapy Response in Head and Neck Squamous CellCarcinoma Based on ceRNA Network and ImmuneInfiltration Analysis

Ya Guo ,1 Wei Kang Pan ,2 Zhong Wei Wang ,1 Wang Hui Su ,1 Kun Xu ,1

Hui Jia ,1 and Jing Chen 1

1Department of Radiation Oncology, The Second Affiliated Hospital, Xi’an Jiao Tong University, Xi’an, 710004 Shaanxi, China2Department of Pediatric Surgery, The Second Affiliated Hospital, Xi’an Jiao Tong University, Xi’an, 710004 Shaanxi, China

Correspondence should be addressed to Ya Guo; [email protected]

Received 3 September 2021; Revised 23 October 2021; Accepted 19 November 2021; Published 6 December 2021

Academic Editor: Rui Liu

Copyright © 2021 Ya Guo et al. This is an open access article distributed under the Creative Commons Attribution License, whichpermits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Objectives. Patients with head and neck squamous cell carcinoma (HNSCC) have poor prognosis and show poor responses toimmune checkpoint (IC) inhibitor (ICI) therapy. Competing endogenous RNA (ceRNA) networks, tumor-infiltrating immunecells (TIICs), and ICIs may influence tumor prognosis and response rates to ICI therapy. This study is aimed at identifyingprognostic and IC-related biomarkers and key TIIC signatures to improve prognosis and ICI therapy response in HNSCCpatients. Methods and Results. Ninety-five long noncoding RNAs (lncRNAs), microRNAs (miRNAs), and 1746 mRNAs wereidentified using three independent methods. We constructed a ceRNA network and estimated the proportions of 22 immunecell subtypes. Ten ceRNAs were related to prognosis according to Kaplan–Meier analysis. Two risk signatures based,respectively, on nine ceRNAs (ANLN, CFL2, ITGA5, KDELC1, KIF23, NFIA, PTX3, RELT, and TMC7) and three immune celltypes (naïve B cells, neutrophils, and regulatory T cells) via univariate Cox regression, least absolute shrinkage and selectionoperator, and multivariate Cox regression analyses could accurately and independently predict the prognosis of HNSCCpatients. Key mRNAs in the ceRNA network were significantly correlated with naïve B cells and regulatory T cells and withstage, grade, and immune and molecular subtype. Eight IC genes exhibited higher expression in tumor tissues and werecorrelated with eight key mRNAs in the ceRNA network in HNSCC patients with different HPV statuses according tocoexpression and TIMER 2.0 analyses. Most drugs were effective in association with expression of these key signatures (ANLN,CFL2, ITGA5, KIF23, NFIA, PTX3, RELT, and TMC7) based on GSCALite analysis. The prognostic value of key biomarkersand associations between key ceRNAs and IC genes were validated using online databases. Eight key ceRNAs were confirmedto predict response to ICI in other cancers based on TIDE analysis. Conclusions. We constructed two risk signatures toaccurately predict prognosis in HNSCC. Key IC-related signatures may be associated with response to ICI therapy.Combinations of ICIs with inhibitors of eight key mRNAs may improve survival outcomes of HNSCC patients.

1. Introduction

Head and neck cancer (HNC) is among the most commonmalignancies worldwide, accounting for about 600,000 newcases annually and 330,000 deaths [1]. Head and neck squa-mous cell carcinoma (HNSCC) is the predominant patho-logical subtype, comprising more than 90% of HNC cases[2]. Despite advances in surgery and radiotherapy, the 5-

year overall survival (OS) rate for HNSCC is still unsatisfac-tory [3, 4]. Thus, there is an indisputably urgent need toidentify effective biomarkers to predict prognosis and OSof patients with HNSCC.

Regulatory competing endogenous RNA (ceRNA) net-works, consisting of long noncoding RNAs (lncRNAs),microRNAs (miRNAs), and mRNAs, have a crucial role inthe processes of tumor occurrence and progression and in

HindawiBioMed Research InternationalVolume 2021, Article ID 4532438, 42 pageshttps://doi.org/10.1155/2021/4532438

https://orcid.org/0000-0002-0343-0976

https://orcid.org/0000-0003-1400-7222

https://orcid.org/0000-0003-3154-5188

https://orcid.org/0000-0002-5704-9679

https://orcid.org/0000-0002-5924-0426

https://orcid.org/0000-0002-9706-227X

https://orcid.org/0000-0001-7846-2537

https://creativecommons.org/licenses/by/4.0/

https://doi.org/10.1155/2021/4532438

prediction of prognosis [5, 6]. Furthermore, accumulatingevidence suggests that immune cell infiltration may have asignificant impact on the prognosis of HNSCC patients [7,8]. Although the prognostic value of ceRNA networks andimmune cell infiltration in HNSCC have been reported,few studies have combined ceRNA networks and immunecell infiltration to predict prognosis in HNSCC patients;moreover, these studies have limitations, and their conclu-sions have been inconsistent. Further, few studies have hadtheir results successfully confirmed in another independentdatabase, and in many, a single analytical method was usedto explore key molecules or immune cells [5, 9–11]. More-over, in HNSCC, multiple clinical characteristics, includinghuman papillomavirus (HPV) status, immune subtype,molecular subtype, grade, and stage, are associated withprognosis [12, 13]. There has not been sufficient systematicevaluation of the association among ceRNA-network RNAs,immune cell infiltration, and these clinical characteristics tofully elucidate the roles of these factors in prognosis. There-fore, a systematic scientific approach is needed to identifyeffective biomarkers for risk assessment of patient prognosis.

HNSCCs comprise a group of highly heterogeneous andimmunosuppressive cancers.

Immunotherapy is aimed at increasing the activity of theimmune system to eliminate cancer cells [14]. Immunecheckpoint (IC) inhibitors (ICIs) represent a broadly effec-tive immunotherapy that can block inhibitory IC pathwaysto restore antitumor immunity.

Anti-PD1/PDL1 ICIs can improve antitumor immuneactivity [4]. Although ICIs exert beneficial therapeutic effectson HNSCC, the response rate to them is still low [15].Therefore, identifying biomarkers able to predict ICI treat-ment response can contribute to patient screening and indi-vidualized treatment approaches and is of great significancefor standardizing immunosuppressive therapy and improv-ing the prognosis of HNSCC patients. Previous studies haverevealed that the expression of immunosuppressive mole-cules and tumor-infiltrating immune cells (TIICs) may reg-ulate the response rate to immunotherapy and prognosis inHNSCC patients. A previous study proposed the epithe-lial–mesenchymal transition CYT Index as a superior pre-dictor of prognosis and immunotherapy response acrossHNSCC [16], and immune cell infiltration scores were iden-tified as sensitive prognostic biomarkers and predictive indi-cators for immunotherapy response [8]. However, noreliable biomarker for predicting response to ICIs and prog-nosis in HNSCC has yet been identified [17, 18].

In this study, mRNA, lncRNA, and miRNA expressionprofiles of HNSCC were downloaded from The CancerGenome Atlas (TCGA) database to construct a ceRNA net-work. CIBERSORT (cell type identification by estimatingrelative subset of known RNA transcripts) was used to esti-mate the abundance of 22 immune cells based on TCGAdataset. Furthermore, two prognostic signatures were devel-oped using a comprehensive bioinformatics method. Next,the association between key genes and immune cells wasevaluated via coexpression analysis. We systematically inves-tigated the associations between key ceRNA signatures andclinical characteristic (including stage, grade, and immune

and molecular subtype). Moreover, IC correlation analysisand key ceRNA-related drug targets were performed toassess the predictive ability of these biomarkers with respectto ICI treatment response in HNSCC and to explore poten-tial targets for improving the effectiveness of immunother-apy. Finally, the prognostic value of key biomarkers andthe associations between key ceRNAs and IC genes were val-idated using online databases.

2. Materials and Methods

2.1. Data Analysis and ceRNA Network Construction. RNAsequencing (RNA-seq) and miRNA-seq data and clinicalinformation for HNSCC patients were extracted from TCGAdatabase (https://gdc-portal.nci.nih.gov/) [19]. HTseq-countprofiles were obtained. In addition, practical extraction andreports Language (Perl) scripts were used to merge all clinicaldata. The R software (version 3.6.3, https://www.r-project.org/) was used to process downloaded files, to convert andeliminate unqualified data, to apply a relatively conservativeapproach to the data analysis, and to identify a suitablenumber of differentially expressed mRNAs (DEmRNAs),miRNAs (DEmiRNAs), and lncRNAs (DElncRNAs). Thelimma, edgeR, and DESeq2 R packages were used for differ-ential expression analysis. Log fold change > 1 or <−1 andfalse discovery rate- (FDR-) adjusted P value < 0.05 wereused as the thresholds to identify significantly expressedbiomarkers. Only biomarkers that were identified usingthree independent analysis methods were selected as thefinal DEmRNAs, DEmiRNAs, and DElncRNAs [20]. AceRNA network was established using the “GDCRNATools”R package. The ceRNA network was evaluated by hypergeo-metric testing and correlation analysis. P < 0:05 was consid-ered to indicate a statistically significant result [21]. Finally,Cytoscape v.3.7.2 was used to visualize the ceRNA net-work [22].

2.2. Construction and Confirmation of ceRNA-RelatedPrognostic Model. We performed univariate Cox regression,least absolute shrinkage and selection operator (LASSO),and multivariate Cox regression analyses to identifyprognosis-related signatures. Moreover, we plotted receiveroperating characteristic (ROC) curves for 1-year, 3-year,and 5-year survival and calculated the corresponding areaunder the curve (AUC) values using the survival and timeROC packages to assess the predictive ability of the ceRNA-related signatures [23]. An AUC of <0.5 was consideredinsignificant. Patients were then divided into high- andlow-risk groups according to the average risk score. We fur-ther calculated survival differences between the high- andlow-risk groups [24]. Next, we performed univariate andmultivariate Cox regression analyses to determine whetherthe risk score based on key ceRNAs was independent of otherclinical characteristics (age, sex, grade, and stage) in the pre-diction of HNSCC prognosis. A hazard ratio ðHRÞ > 1 andP < 0:05 were considered to indicate unfavorable prognosticfactors, whereas HR < 1 and P < 0:05 indicated favorableprognostic factors [25].

2 BioMed Research International

https://gdc-portal.nci.nih.gov/

https://www.r-project.org/


0

500

1000

1500

2000Pr

otei

n co

ding

Long

non

-cod

ing IG

Pseu

doge

ne

TEC

Oth

er n

cRN

A

No.

of d

iffer

entia

lly ex

pres

sed

gene

s

Up-regulated

Down-regulated

(a)

0

500

1000

1500

2000

Prot

ein

codi

ng

Long

non

-cod

ing IG

Pseu

doge

ne

TEC

Oth

er n

cRN

A

No.

of d

iffer

entia

lly ex

pres

sed

gene

s

Up-regulated

Down-regulated

(b)

Figure 1: Continued.

3BioMed Research International

0

500

1000

1500

2000

Prot

ein

codi

ng

Long

non

-cod

ing

Pseu

doge

ne IG

TEC

Oth

er n

cRN

A

No.

of d

iffer

entia

lly ex

pres

sed

gene

s

Up-regulated

Down-regulated

(c)

Size of each list

0

57.5

115115

Deseq2-LnRNA

115

edgeR-LnRNA

109

limma-LnRNA

Number of elements: specific (1) or shared by 2, 3, ... lists

95

3

20

2

14

1

0 0

14

20

0 0

95

Deseq2-LnRNA edgeR-LnRNA

limma-LnRNA

(d)



Size of each list

0

182

91166

Deseq2-miRNA

182

edgeR-miRNA

169

limma-miRNA


146

3

28

2

23

1

2 13

8

13

5 10

146

Deseq2-miRNA edgeR-miRNA

limma-miRNA

(e)

Size of each list

0

1110.5

2221 2221

Deseq2-mRNA

2213

edgeR-mRNA

2069

limma-mRNA


1746

3

460

2

345

1

21 23

301

438

16 6

1746

Deseq2-mRNA edgeR-mRNA

limma-mRNA

(f)



2.3. Construction of Immune Cell-Related Prognostic Model.The CIBERSORT algorithm was used to estimate the abun-dance of 22 immune cell subtypes according to the RNA-seq count data [26]. The associations between immune celltypes and prognosis were investigated by univariate Coxregression, LASSO regression, and multivariate Cox regres-sion. We further assessed the sensitivity and specificity ofimmune cell-related prognostic models using ROC curves.In addition, we performed an independent prognostic anal-ysis to assess whether the risk score based on key immunecells could predict HNSCC prognosis independently ofother clinical characteristics.

2.4. Associations between Key ceRNAs and SignificantImmune Cell Subtypes. The associations between significantmRNAs in the ceRNA network and key immune cells wereinvestigated using Pearson correlation coefficients. More-over, we analyzed the associations between the gene riskscore and key immune cells.

2.5. Clinical Correlation Analysis. The Tumor Immune Sys-tem Interactions Database (TISIDB; http://cis.hku.hk/TISIDB) was used to explore Spearman correlationsbetween the identified key ceRNAs and clinical character-istics. Moreover, the expression distribution patterns ofnine key ceRNAs across immune and molecular subtypeswere determined using TISIDB [27]. The immune sub-types comprised five groups: C1 (wound healing), C2(IFN-gamma dominant), C3 (inflammatory), C4 (lympho-cyte depleted), and C6 (TGF-β dominant) [12]. HNSCCcan be classified into four molecular subtypes, namely,atypical (AT), basal (BA), classical (CL), and mesenchymal(MS) [13].

2.6. Correlations between Immunotherapy Response and KeyceRNAs. The expression of ICs is correlated with theimmune response to immunotherapy. We extracted theexpression levels of eight immune-checkpoint genes(CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2,

SYNM

SYBU

ACSL1

SASH1

PRKAA2

DIO2

CFL2

RASD1

hsa-miR-130b-3p

LPL

hsa-miR-454-3p

PRUNE2

NFIX

SCARA3

TACC2

NFIA

GPT2

GADD45B

H19

ZFP36

PDGFRB

LOXL2

KLF4KDELC1

hsa-miR-29a-3p

SOX12

FOXO4

NID2ZBTB47

AMOTSPRY1

C1RL-AS1

HOXA10CDCA4

GNA12CBX2

EN2 CHEK1

RELTMYBL1

LAMC1

COL12A1

DCBLD2P3H2

KIF23

hsa-miR-424-5p

SH2D2APTHLHCCNE1

SPRY4STX1A

RFLNBCOL5A2

PI15ITGA11

BMP1

PDK4

C1QTNF6

EMP1PTX3

ADAMTS7

HELLShsa-miR-29c-3p

COL7A1 hsa-miR-7-5p

TFRC

ITGA5

KIRREL1

hsa-miR-338-3p

SPARC

VAV2

KCNQ1OT1

hsa-miR-335-5p

TMC7

ANLN

HAS2

CDC25B

APLNHOXC11

SCN8A

(g)

Figure 1: Identification of DElncRNAs, DEmiRNAs, DEmRNAs, and construction of the ceRNA network. Differentially expressed geneswere identified by three independent methods. (a) Deseq2, (b) edgeR, and (c) limma. (d–f) Venn diagram displaying the intersection of(d) DElncRNAs, (e) DEmiRNAs, and (f) DEmRNAs. (g) Construction of the ceRNA network. Red circles represent miRNAs, greencircles represent lncRNAs, and purple circles represent mRNAs. TEC: to be experimentally confirmed; IG: immunoglobulin. Log foldchange > 1 or <−1 and FDR-adjusted P value < 0.05 were used to identify significantly expressed biomarkers.


http://cis.hku.hk/TISIDB

http://cis.hku.hk/TISIDB

++++++

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++++++++

+++

+ +

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++++++++++

+ + ++

p = 0.011

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

ACSL1+

+

High

Low

246 182 108 70 43 22 15 12 8 5 4 3 3 3 2 1 0 0 0 0 0

247 191 106 63 49 30 19 14 8 8 7 6 4 3 2 1 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

ACSL

1

(a)

+++++++++++++

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ +++

+ + + +

+

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++++++ + +

+++ +

+p = 0.026

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 178 100 63 43 27 19 14 7 7 6 5 4 4 2 0 0 0 0 0 0

247 195 114 70 49 25 15 12 9 6 5 4 3 2 2 2 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

CDCA

4

CDCA4+

+

High

Low

(b)



++++++++

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++

++ +

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++++++ + +

++++ ++

p = 0.001

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 186 110 61 39 24 15 11 5 4 3 3 3 2 2 1 0 0 0 0 0

247 187 104 72 53 28 19 15 11 9 8 6 4 4 2 1 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

GN

A12

GNA12+

+

High

Low

(c)

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++ + ++++++ + ++

+++++++++++

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++++++++++

+

+ +

p = 0.046

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 189 114 71 50 25 15 15 9 7 6 5 3 2 1 1 1 1 0 0 0

247 184 100 62 42 27 19 11 7 6 5 4 4 4 3 1 0 0 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

hsa−

miR

−29c

−3p

hsa−miR−29c−3p+

+

High

Low

(d)



++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++ ++++

++ +

+++++++

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+ + + + +

+ +p = 0.038

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 177 98 59 44 21 16 14 6 4 3 2 1 0 0 0 0 0 0 0 0

247 196 116 74 48 31 18 12 10 9 8 7 6 6 4 2 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

hsa−

miR

−130

b−3p

hsa−miR−130b−3p+

+

High

Low

(e)

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++++++++ ++ +

++ ++ +

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++ + ++

++ +

p = 0.002

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 181 104 65 42 30 20 14 7 5 4 4 3 2 2 1 0 0 0 0 0

247 192 110 68 50 22 14 12 9 8 7 5 4 4 2 1 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

ITG

A5

ITGA5+

+

High

Low

(f)



++++++++

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+ +

+

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++ + +++++ ++ + ++

+p = 0.003

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 182 99 56 33 21 14 8 6 5 3 2 2 1 1 0 0 0 0 0 0

247 191 115 77 59 31 20 18 10 8 8 7 5 5 3 2 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

KDEL

C1

KDELC1+

+

High

Low

(g)

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++++++++ ++ ++

+

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++ +++ +

++ ++ ++

+p = 0.047

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 185 105 70 50 23 13 8 5 3 2 2 2 1 1 1 0 0 0 0 0

247 188 109 63 42 29 21 18 11 10 9 7 5 5 3 1 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

PRKA

A2

PRKAA2+

+

High

Low

(h)



+++++++++++++++++++++++++++++++++

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++

++

+

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+ ++ +++

+ + ++ +

+

p = 0.008

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 174 107 70 46 27 18 12 8 6 5 5 4 3 2 2 1 1 0 0 0

247 199 107 63 46 25 16 14 8 7 6 4 3 3 2 0 0 0 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

PRU

NE2

PRUNE2+

+

High

Low

(i)

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ +

+

+ +

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++++ +++ ++ ++ +

+ +p = 0.004

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

246 179 95 60 41 24 14 10 6 5 3 3 2 1 0 0 0 0 0 0 0

247 194 119 73 51 28 20 16 10 8 8 6 5 5 4 2 1 1 0 0 0Low

High

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

PTX3

PTX3+

+

High

Low

(j)

Figure 2: Survival analysis of key members in the ceRNA network. (a–j) K-M survival curves for the significant members of the ceRNAnetwork. (a) ACSL1, (b) CDCA4, (c) GNA12, (d) hsa-miR-29c-3p, (e) hsa-miR-130b-3p, (f) ITGA5, (g) KDELC1, (h) PRKAA2, (i)PRUNE2, and (j) PTX3.


TIGIT, and SIGLEC15) and compared their differences inexpression between tumor and normal tissues in HNSCCusing the R packages limma, ggplot2, and pheatmap. Coex-pression analysis was used to define the correlations betweenthe expression levels of nine key ceRNAs and those of theeight immune-checkpoint genes. A two-gene correlationmap was obtained using the R package ggstatsplot. A P valueof less than 0.05 was considered to indicate statistical signif-icance. The Tumor Immune Dysfunction and Exclusion(TIDE, http://tide.dfci.harvard.edu) algorithm was then usedto estimate the predictive power of the custom biomarkerwith respect to response outcome and OS [28]. We con-firmed the correlations between the eight key mRNAs andresponse to ICIs in other cancers using TIDE.

2.7. Correlations between Expression of Key ceRNAs andDrug Sensitivity. Gene Set Cancer Analysis (GSCALite,http://bioinfo.life.hust.edu.cn/web/GSCALite/) is a webserver for dynamic analysis and visualization of gene setsin cancer and drug sensitivity correlations. The expressionof each gene in the gene set was determined, and small mol-ecule/drug sensitivity (half-maximal inhibitory concentra-tion; IC50) was obtained using Spearman’s correlationanalysis. Correlations with FDR < 0:05 were considered torepresent results [29–31]. We used GSCALite to analyzethe correlations between drug sensitivity and the signatureof nine key ceRNAs.

2.8. Validation of Results. LOGpc (Long-term Outcome andGene Expression Profiling Database of pan-cancers, http://bioinfo.henu.edu.cn/DatabaseList.jsp) includes 209 geneexpression datasets and provides 13 types of survival termsfor 31,310 patients with 27 distinct malignancies [32]. Wefirst confirmed the prognostic value of major target genesusing LOGpc datasets. Next, we confirmed the prognosticvalue of key immune cell types and ceRNAs using theTIMER 2.0 database [33]. Finally, we confirmed the expres-sion of eight immune-checkpoint genes and determined thecorrelations between the expression levels of these genes and

nine key genes in HNSCC with different HPV statuses usingthe gene correlation TIMER 2.0 module (http://timer.cistrome.org/) [34].

2.9. Statistical Analysis. All statistical analyses were carriedout using R (version 3.6.3, https://www.r-project.org/) andthe limma, GDCRNATools, ggplot2, rms, glmnet, survmi-ner, and timeROC packages. A two-sided P value of less than0.05 was considered to indicate statistical significance.

3. Results

3.1. Identification of Differentially Expressed Genes,DEmiRNAs, and DElncRNAs. We identified 115 DElncR-NAs, 166 DEmiRNAs, and 2221 DEmRNAs using theDESeq2 package in the R programming language. Of these,115 lncRNAs, 182 miRNAs, and 2213 mRNAs were foundto be differentially expressed in HNSCC using the edgeRmethod. Next, we acquired 109 DElncRNAs, 169 DEmiR-NAs, and 2069 DEmRNAs using the R package limma. Atotal of 95 DElncRNAs, 146 DEmiRNAs, and 1746 DEmR-NAs were identified using three independent methods(Figures 1(a)–1(f)).

3.2. Construction of ceRNA Network and Survival Analysis.We constructed a lncRNA–miRNA–mRNA ceRNA net-work, which included three lncRNAs, eight miRNAs, and69 mRNAs (Figure 1(g)). Then, Kaplan–Meier (K-M) sur-vival analysis was performed to identify prognostic membersin the constructed ceRNA network. Our results showed that10 key markers, including ACSL1, CDCA4, GNA12, hsa-miR-29c-3p, hsa-miR-130b-3p, ITGA5, KDELC1, PRKAA2,PRUNE2, and PTX3, were significantly correlated with sur-vival (Figures 2(a)–2(j)).

3.3. Construction of ceRNA-Related Prognostic Model. Thir-teen ceRNAs were identified as prognosis-related signaturesusing univariate Cox regression (Table 1). LASSO regressionanalysis revealed that nine key markers were essential formodeling (Figures 3(a) and 3(b)); these were subjected to

Table 1: Identification of prognosis-related ceRNAs by univariate Cox regression analysis.

Id HR HR.95L HR.95H P value

ANLN 1.2194 1.0487 1.4179 0.0099

C1QTNF6 1.1279 1.0083 1.2617 0.0354

CFL2 1.2049 1.0554 1.3756 0.0058

DCBLD2 1.1560 1.0149 1.3166 0.0290

GNA12 1.3001 1.0687 1.5816 0.0087

ITGA5 1.2156 1.0892 1.3567 0.0005

KDELC1 1.3339 1.1295 1.5753 0.0007

KIF23 1.2437 1.0289 1.5034 0.0242

NFIA 0.7776 0.6488 0.9319 0.0065

PRUNE2 1.0667 1.0078 1.1292 0.0260

PTX3 1.1174 1.0509 1.1882 0.0004

RELT 1.2274 1.0165 1.4822 0.0332

TMC7 1.1395 1.0046 1.2926 0.0422


http://tide.dfci.harvard.edu

http://bioinfo.life.hust.edu.cn/web/GSCALite/

http://bioinfo.henu.edu.cn/DatabaseList.jsp

http://bioinfo.henu.edu.cn/DatabaseList.jsp

http://timer.cistrome.org/



−7 −6 −5 −4 −3

−0.20

−0.15

−0.10

−0.05

0.00

0.05

0.10

0.15

Log lambda

Coef

ficie

nts

13 13 12 9 7

1

2

3

4

5

67

8

9

10

11

12

13

(a)

−7 −6 −5 −4 −3

12.15

12.20

12.25

12.30

12.35

12.40

12.45

Log (𝜆)

Part

ial l

ikel

ihoo

d de

vian

ce

13 13 13 13 13 13 11 9 9 9 8 8 7 5 3

(b)



TMC7

RELT

PTX3

NFIA

KIF23

KDELC1

ITGA5

CFL2

ANLN

(N = 493)

(N = 493)

(N = 493)

(N = 493)

(N = 493)

(N = 493)

(N = 493)

(N = 493)

(N = 493)

1.05(0.90 − 1.22)

1.13(0.91 − 1.40)

1.08(1.00 − 1.16)

0.82(0.67 − 0.99)

1.16(0.88 − 1.54)

1.08(0.87 − 1.35)

1.01(0.86 − 1.19)

1.10(0.94 − 1.28)

1.04(0.81 − 1.33)

0.567

0.267

0.046 ⁎

0.044 ⁎

0.301

0.466

0.858

0.234

0.76

# Events: 215; global p-value (log-rank): 0.00029134 AIC: 2320.19; concordance index: 0.63

0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6

Hazard ratio

(c)

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

1 − specificity

Sens

itivi

ty

AUC at 1 years: 0.633AUC at 3 years: 0.681AUC at 5 years: 0.591

(d)



++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++ +

+

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++++++ ++++ +++ ++ ++

+p < 0.001

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

RiskHigh risk

Low risk

246169 86 50 30 17 9 6 4 3 2 2 2 1 1 0 0 0 0 0 0247204128 83 62 35 25 20 12 10 9 7 5 5 3 2 1 1 0 0 0Low risk

High risk

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Risk

+

+

(e)

Age

Gender

Grade

Stage

T

N

Risk score

0.220

0.102

0.049

0.003

0.004

<0.001

<0.001

pvalue

1.014 (0.991 − 1.038)

0.622 (0.352 − 1.099)

1.517 (1.002 − 2.298)

1.967 (1.263 − 3.063)

1.518 (1.141 − 2.018)

1.827 (1.362 − 2.451)

2.665 (1.606 − 4.422)

Hazard ratio

Hazard ratio0 1 2 3 4

(f)



multivariable model analysis. Only two key biomarkers(NFIA and PTX3) had a significant impact on HNSCC prog-nosis (Figure 3(c)). In addition, analysis of the ROC curvesindicated that the AUCs for 1-, 3-, and 5-year OS were0.633, 0.681, and 0.591, respectively (Figure 3(d)). We fur-ther plotted survival curves for the high- and low-riskgroups. As shown in Figure 3(e), patients in the low-riskgroup had significantly longer OS probability than those inthe high-risk group. Finally, univariate and multivariateCox regression analyses demonstrated that the risk scorebased on key ceRNAs was an independent predictor of poorprognosis in HNSCC patients (Figures 3(f) and 3(g)).

3.4. Immune Cell Infiltration Analysis. CIBERSORTwas usedto assess immune cell infiltration in each HNSCC sample.The results revealed significant differences in the proportionsof immune cell infiltration (Figure 4(a)). In addition, the gen-erated heatmap showed that 11 immune cell subtypes, M1macrophages, CD8 T cells, M0 macrophages, M2 macro-phages, resting CD4 memory T cells, memory B cell, naïveB cells, and regulatory T cells (Treg), monocytes, restingmyeloid dendritic cells, and activated mast cells presentedsignificant different proportions between the tumor groupand the normal group (Figure 4(b)).

3.5. Construction and Validation of Immune Cell-RelatedPrognostic Model. Univariate Cox regression, LASSO regres-sion, multivariate Cox regression, and independent prognos-tic analysis were used to analyze the associations betweendifferent immune cell subtypes and prognosis. The resultsshowed that three variables were related to prognosis inHNSCC: high infiltration of Tregs and naïve B cells wasassociated with a favorable prognosis, whereas increasedlevels of neutrophils were correlated with worse prognosis(Figures 4(c)–4(e)). According to ROC curve analysis, the

AUCs of the 1-, 3-, and 5-year prognosis models were0.625, 0.626, and 0.568, respectively (Figure 4(f)). The K-Mcurve analysis indicated that the high-risk group had anunfavorable prognosis (Figure 4(g)). Finally, we found thatthe immune cell-related risk score was an independentfactor for predicting prognosis in HNSCC (Figures 4(h)and 4(i)).

3.6. Relationships between Key ceRNAs and SignificantImmune Cell Signatures. There were significant correlationsbetween key molecules in the ceRNA network and immunecell signatures (Figure 5(a)). The coexpression analysisresults indicated that CFL2, ITGA5, KDELC1, and TMC7were negatively associated with levels of naïve B cell infiltra-tion (Figures 5(b)–5(e)), whereas ANLN, ITGA5, KIF23,and TMC7 expression was negatively associated with Treglevels (Figures 5(f)–5(i)). As shown in Figures 5(j)–5(l), sig-nificant immune cell signatures in different risk score exhib-ited statistical significance, and higher levels of naïve B cellswere associated with lower risk scores (P = 0:0085). A simi-lar result was observed for Tregs (P = 4e − 9).

3.7. Clinical Correlation Analysis. We evaluated the correla-tions of the nine key signatures with tumor grade and stage.Our results showed that increased expression of ANLN andKIF23 was associated with higher tumor grade (Figures 6(a)and 6(b)). Moreover, the expression levels of ITGA5,KDELC1, and PTX3 were positively correlated with the stageof HNSCC, whereas NFIA expression decreased at higherHNSCC stages (Figures 6(c)–6(f)). Furthermore, our resultsshowed that ANLN, KDELC1, KIF23, and NFIA expressionwas associated with different immune subtypes. ANLN andKIF23 showed increased expression in subtypes C1 andC2. KDELC1 showed higher expression in the C1, C2, andC6 subtypes, indicating that it may be mainly associated

Age

Gender

Grade

Stage

T

N

Risk score

0.824

0.137

0.093

0.793

0.039

0.002

0.004

pvalue

1.003 (0.978 − 1.028)

0.640 (0.355 − 1.153)

1.497 (0.936 − 2.395)

0.914 (0.465 − 1.794)

1.562 (1.023 − 2.386)

1.717 (1.213 − 2.430)

2.184 (1.281 − 3.724)

Hazard ratio

Hazard ratio0.0 1.0 2.0 3.0

(g)

Figure 3: Development and validation of the ceRNA-related prognostic model. (a–c) LASSO and multivariate Cox regression analyses wereapplied to investigate the correlation between OS and ceRNAs. (d) ROC analysis showing the accuracy of the prediction model. (e) K-Msurvival curves comparing the high-risk and low-risk groups and the demonstrating predictive ability of our model. (f, g) Assessment ofthe independence of ceRNA-related prognostic model through (f) univariate and (g) multivariate Cox regression analyses. HR > 1 and Pvalue < 0.05 indicate poor prognostic factors. HR < 1 and P values < 0.05 indicate favorable prognostic factors.


100%

50%

Perc

ent

0%Type

B cell memory

B cell naive

B cell plasma

Eosinophil

Macrophage M0

Macrophage M1

Macrophage M2

Mast cell activated

Mast cell resting

Monocyte

Myeloid dendritic cell activated

Myeloid dendritic cell resting

Neutrophil

NK cell activated

NK cell resting

T cell CD4+ memory activated

T cell CD4+ memory resting

T cell CD4+ naive

T cell CD8+

T cell follicular helper

T cell gamma delta

T cell regulatory (tregs)

(a)



GroupsMacrophage M1⁎⁎⁎ 9.00e–07 2

–2

5.83e–039.04e–025.49e–017.07e–018.02e–212.64e–015.31e–019.09e–021.25e–012.48e–029.29e–072.37e–034.53e–031.52e–033.90e–011.48e–011.31e–012.79e–147.94e–019.64e–073.66e–07

T cell CD8+⁎⁎T cell CD4+ memory activatedT cell follicular helperNK cell activatedMacrophage M0⁎⁎⁎B cell plasmaMyeloid dendritic cell activatedNK cell restingMast cell restingMacrophage M2⁎T cell CD4+ memory resting⁎⁎⁎B cell memory⁎⁎T cell regulatory (tregs)⁎⁎B cell naive⁎⁎T cell gamma deltaT cell CD4+ naiveEosinophilMonocyte⁎⁎⁎NeutrophilMyeloid dendritic cell resting⁎⁎⁎Mast cell activated⁎⁎⁎

1

0

–1

GroupsTumorNormal

(b)

−6 −5 −4 −3

−15

−10

−5

0

5

10

15

20

Log lambda

Coeffi

cien

ts

5 5 5 5

1

2

3

4

5

(c)



−6 −5 −4 −3

11.7

11.8

11.9

12.0

Log (𝜆)

Part

ial l

ikel

ihoo

d de

vian

ce

5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 3 2 0

(d)

Neutrophils

T cells regulatory (tregs)

B cells naive

(N= 338)

(N= 338)

(N= 338)

6.2e+03(2.5e+00 − 1.5e+07)

7.9e−08(4.1e−15 − 1.5e+00)

4.6e−02(2.4e−03 − 8.9e−01)

0.029⁎

0.056

0.042⁎

# Events: 144; global p-value (log-rank): 0.00099182AIC: 1438.2; concordance index: 0.61

1e−16 1e−12 1e−08 0.0001 1 10000 1e+08

Hazard ratio

(e)



0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

1 − specificity

Sens

itivi

ty

AUC at 1 years: 0.625AUC at 3 years: 0.626AUC at 5 years: 0.568

(f)

++++

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++ +

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

+++++ ++++

++ +

p = 0.015

0.00

0.25

0.50

0.75

1.00

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Surv

ival

pro

babi

lity

169126 64 42 28 15 7 5 5 4 3 3 3 2 2 1 0 0 0 0 0169134 87 56 37 21 14 11 7 6 5 4 3 3 1 1 1 1 0 0 0Low risk

High risk

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Time (years)

Risk

RiskHigh risk

Low risk

+

+

(g)



with worse survival. NFIA showed higher expression in sub-type C3 and decreased expression in C4, which predictedbetter survival (Figures 6(g)–6(i)). CFL2, ITGA5, KDELC1,and PTX3 were highly expressed in the MS subtype.Increased expression of RLET and TMC7 was mainly foundin the basal subtype, and higher expression of NFIA wasassociated with the AT subtype (Figures 6(k)–6(q)).

3.8. Key ceRNA Expression Was Correlated withImmunotherapy Response and Drug Sensitivity. Multiple ICgenes including CD274, CTLA4, HAVCR2, LAG3, PDCD1,SIGLEC15, TIGIT, and TIM3 exhibited higher expression intumor tissues compared with normal tissues (Figure 7(a)).The expression of ANLN, ITGA5, and KIF23 was positivelyassociated with the expression of CD274 (PDL1), HAVCR2,and SIGLEC15. CFL2 and RELT expression was positivelycorrelated with the expression of CD274 (PDL1), CTLA4,HAVCR2, LAG3, PDCD1 (PD1), SIGLEC15, and TIGIT.

NFIA was positively related to the abovementioned sevenimmunosuppressive genes, except for CD274. Overexpressionof PTX3 was associated with increased CTLA4, HAVCR2,SIGLEC15, and TIGIT expression. TMC7 expression wasnegatively correlated with the expression of CTLA4,HAVCR2, LAG3, PDCD1 (PD1), and TIGIT (Figure 7(b)).

In addition, our biomarker could effectively predict anti-PD1 response compared with published biomarkers (MSI,CD274, CD8, IFNG, etc.) (Figure 7(c)). Moreover, Spear-man’s correlation analysis was performed to explore the cor-relations of the expression of the nine key genes with drugsensitivity in terms of IC50 values. Our results showed thatmost drugs were effective in association with increasedexpression of CFL2, TMC7, PTX3, ANLN, NFIA, andKIF23, whereas RELT and ITGA5 were negatively regulatedby most drugs. Specifically, these molecules could beexploited as potential therapeutic drug targets for HNSCC(Figure 7(d)).

Age

Gender

Grade

Stage

T

N

Risk score

0.670

0.191

0.125

0.025

0.010

<0.001

0.029

pvalue

1.006 (0.979 − 1.034)

0.637 (0.325 − 1.251)

1.464 (0.900 − 2.383)

1.715 (1.071 − 2.747)

1.556 (1.111 − 2.179)

1.918 (1.346 − 2.734)

1.738 (1.057 − 2.858)

Hazard ratio

Hazard ratio0.0 0.5 1.0 1.5 2.0 2.5

(h)

Age

Gender

Grade

Stage

T

N

Risk score

0.762

0.102

0.011

0.054

0.005

<0.001

0.006

pvalue

0.996 (0.968 − 1.024)

0.562 (0.281 − 1.122)

2.183 (1.194 − 3.992)

0.466 (0.215 − 1.012)

2.171 (1.261 − 3.735)

2.543 (1.631 − 3.964)

2.398 (1.281 − 4.488)

Hazard ratio

Hazard ratio0 1 2 3 4

(i)

Figure 4: Construction and validation of the immune cells related prognostic model. (a) Distribution of 22 immune cell subtypes estimatedby CIBERSORT in HNSCC. (b) Heatmap showing the levels of 22 immune cell subtypes in tumor and normal samples. (c–e) LASSO andmultivariate Cox regression were used to investigate the correlation between OS and immune cell infiltration. (f, g) ROC and Kaplan–Meiersurvival curves showing the accuracy of immune cell subtypes and related prognostic models for predicting OS. (h) Univariate analysis and(i) multivariate Cox regression analysis confirmed the independence of the immune cell-related prognostic model.


−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1A

NLN

CFL2

ITG

A5

KDEL

C1

KIF2

3

NFI

A

PTX3

RELT

TMC7

B ce

lls n

aive

T ce

lls re

gulat

ory

(treg

s)

Neu

troph

ils

ANLN

CFL2

ITGA5

KDELC1

KIF23

NFIA

PTX3

RELT

TMC7

B cells naive

T cells regulatory (tregs)

Neutrophils

1

0.26

0.32

0.24

0.75

0.11

0.09

0.28

0.25

−0.1

−0.25

0.05

0.26

1

0.43

0.3

0.15

−0.05

0.26

0.11

0.11

−0.24

−0.14

0

0.32

0.43

1

0.56

0.11

−0.14

0.41

0.32

0.48

−0.33

−0.29

0.04

0.24

0.3

0.56

1

0.14

−0.22

0.41

0.15

0.32

−0.18

−0.17

0.03

0.75

0.15

0.11

0.14

1

0.03

−0.04

0.19

0.09

−0.06

−0.22

−0.01

0.11

−0.05

−0.14

−0.22

0.03

1

−0.05

−0.05

−0.01

0.16

0.11

−0.05

0.09

0.26

0.41

0.41

−0.04

−0.05

1

0.19

0.07

−0.01

−0.12

0.04

0.28

0.11

0.32

0.15

0.19

−0.05

0.19

1

0.29

−0.12

−0.13

0.11

0.25

0.11

0.48

0.32

0.09

−0.01

0.07

0.29

1

−0.21

−0.29

0.14

−0.1

−0.24

−0.33

−0.18

−0.06

0.16

−0.01

−0.12

−0.21

1

0.23

0.04

−0.25

−0.14

−0.29

−0.17

−0.22

0.11

−0.12

−0.13

−0.29

0.23

1

−0.08

0.05

0

0.04

0.03

−0.01

−0.05

0.04

0.11

0.14

0.04

−0.08

1

(a)



R = −0.24, p = 1.1e−05

0.0

0.1

0.2

0.3

2 4 6CFL2

B ce

lls n

aive

(b)

R = −0.33, p = 9e−10

0.0

0.1

0.2

0.3

4 6 8 10ITGA5

B ce

lls n

aive

(c)



R = −0.18, p = 0.00099

0.0

0.1

0.2

0.3

0 2 4KDELC1

B ce

lls n

aive

(d)

R = −0.21, p = 8e−05

0.0

0.1

0.2

0.3

0 2 4TMC7

B ce

lls n

aive

(e)



R = −0.25, p = 4.1e−06

0.00

0.02

0.04

0.06

0.08

3 4 5 6 7 8ANLN

T ce

lls re

gulat

ory

(treg

s)

(f)

R = −0.29, p = 6.3e−08

0.00

0.02

0.04

0.06

0.08

4 6 8 10ITGA5

T ce

lls re

gulat

ory

(treg

s)

(g)



R = −0.22, p = 6.3e−05

0.00

0.02

0.04

0.06

0.08

3 4 5 6KIF23

T ce

lls re

gulat

ory

(treg

s)

(h)

R = −0.29, p = 5.7e−08

0.00

0.02

0.04

0.06

0.08

0 2 4TMC7

T ce

lls re

gulat

ory

(treg

s)

(i)



0.0085

0.0

0.1

0.2

0.3

High risk Low riskRisk

B ce

lls n

aive

RiskHigh risk

Low risk

(j)

0.22

0.00

0.05

0.10

0.15


Neu

trop

hils

RiskHigh risk

Low risk

(k)



3.9. Validation of the Prognostic Value of Biomarkers andAssociations between Key ceRNAs and IC Genes. LOGpcwas used to perform survival analysis for key ceRNAs. Theresults indicated that higher expression levels of CFL2,ITGA5, KDELC1, PTX3, and RELT were correlated withpoor prognosis in HNSCC patients, whereas increasedexpression of NFIA was associated with longer OS. Therewas no correlation between ANLN, KIF23, or TMC7 expres-sion and prognosis in HNSCC patients (Table 2). The asso-ciations between pivotal ceRNAs and prognosis were thendetermined using TIMER 2.0. We found that high expres-sion levels of ANLN, CFL2, ITGA5, KDELC1, PTX3, RELT,and TMC7 were significantly associated with shorter OS inHNSCC patients, whereas higher expression of NFIA wascorrelated with longer OS. In HPV-positive HNSCCpatients, high expression of ITGA5, KDELC1, and TMC7was associated with poor prognosis, whereas high NFIAexpression was correlated with good prognosis. ANLN,CFL2, KDELC1, PTX3, and RELT were associated with unfa-vorable prognosis in HPV-negative HNSCC (Figure 8(a)). Inaddition, the prognostic roles of immune cells in HNSCCpatients with different HPV statuses were confirmed usingthe TIMER 2.0 database. In both HPV-positive and HPV-negative HNSCC patients, high infiltration levels of naïveB cells and Tregs were associated with a favorable progno-sis according to several algorithms, whereas high neutro-phil infiltration indicated an unfavorable prognosis(Figures 8(b)–8(d)). Eight tumor-related immunosuppres-

sive molecules, CD274, CTLA4, HAVCR2, LAG3, PDCD1(PD1), PDCD1LG2, SIGLEC15, and TIGIT, had higherexpression in HNSCC tumor tissues compared with HNSCCnormal tissues.

Notably, most IC molecules showed significantlyincreased expression in HPV-positive HNSCC patients(Figure 8(e)). We also found that the expression of theseICI factors was significantly positively correlated with theexpression of seven key mRNAs (ANLN, CFL2, ITGA5,KIF23, NFIA, PTX3, and RELT), whereas the expression ofimmunosuppressive molecules was negatively associatedwith TMC7 expression, especially in HPV-positive HNSCCpatients (Figure 8(f)). We further discovered that eight keyceRNAs were correlated with response to ICI in othercancers (melanoma, bladder, kidney, and glioblastoma)(Table 3).

These results are consistent with the results of our study.

4. Discussion

HNSCC is among the deadliest malignancies in humans anda significant cause of cancer-related deaths worldwide [35].Despite advances in the screening, diagnosis, and treatmentof HNSCC in recent decades, especially in terms of immuno-therapy, the prognosis of HNSCC patients remains verypoor [36]. Recent studies have explored the roles of immunecell infiltration and ceRNA networks in HNSCC. However,these studies have failed to accurately identify key molecules

4e−09

0.00

0.02

0.04

0.06

0.08


T ce

lls re

gulat

ory

(treg

s)

RiskHigh risk

Low risk

(l)

Figure 5: Analysis of the correlations of significant genes in the ceRNA network and gene risk score with immune cell infiltration. (a)Heatmap displaying correlations between prognostic immune cells and key genes in the ceRNA network. (b–e) Expression of CFL2,ITGA5, KDELC1, and TMC7 was negatively associated with infiltration of naïve B cells. (f–i) ANLN, ITGA5, KIF23, and TMC7expression was negatively associated with Tregs. (j–l) Pearson’s correlation analysis was performed to evaluate the associations betweenkey immune cells and gene risk score.


3

G1

G2

G3

G4

Gra

de

4 5 6ANLN

7 8

(a)

2.5

G1

G2

G3

G4

Gra

de3.0 3.5

KIF236.56.05.55.04.54.0

(b)

9 10

Stage I

Stage II

Stage III

Stage IV

Stag

e

4 5 6ITGA5

7 8

(c)



5

Stage I

Stage II

Stage III

Stage IV

Stag

e

0 1 2KDELC1

3 4

(d)

6

Stage I

Stage II

Stage III

Stage IV

Stag

e

1 2 3NFIA

4 5

(e)



4 6

Stage I

Stage II

Stage III

Stage IV

Stag

e

–6 –4 –2PTX3

0 2

(f)

8

6

4

Expr

essio

n (lo

g2CP

M)

2

C1 C2 C3

HNSC :: ANLN_expPv = 3.51e–02

n = C1 128, C2 379, C3 2, C4 2, C6 3

SubtypeC4 C6

(g)

8

6

4

Expr

essio

n (lo

g2CP

M)

2

C1 C2 C3


n = C1 128, C2 379, C3 2, C4 2, C6 3

SubtypeC4 C6

(h)

8

6

4

Expr

essio

n (lo

g2CP

M)

2

C1 C2 C3


n = C1 128, C2 379, C3 2, C4 2, C6 3

SubtypeC4 C6

(i)

6

8

4

Expr

essio

n (lo

g2CP

M)

2

C1 C2 C3

HNSC :: NFIA_expPv = 3.78e–03

n = C1 128, C2 379, C3 2, C4 2, C6 3

SubtypeC4 C6

(j)

8

6

4

Expr

essio

n (lo

g2CP

M)

2

C1 C2 C3


n = C1 128, C2 379, C3 2, C4 2, C6 3

SubtypeC4 C6

(k)



7.5

10.0

5.0

Expr

essio

n (lo

g2CP

M)

2.5

Atyp

ical

HNSC :: ITGA5_expPv = 2.24e–23n = atypical 67,

basal 87,classical 48,

mesenchymal 74

Subtype

Basa

l

Clas

sical

Mes

ench

ymal

(l)

7.5

10.0

5.0

Expr

essio

n (lo

g2CP

M)

2.5

Atyp

ical

HNSC :: ITGA5_expPv = 2.24e–23n = atypical 67,


mesenchymal 74

Subtype

Basa

l

Clas

sical

Mes

ench

ymal

(m)

4

6

Expr

essio

n (lo

g2CP

M)

2

Atyp

ical

HNSC :: NFIA_expPv = 3.43e–07n = atypical 67,


mesenchymal 74

Subtype

Basa

l

Clas

sical

Mes

ench

ymal

(n)

0

5

Expr

essio

n (lo

g2CP

M)

–5

Atyp

ical

HNSC :: PTX3_expPv = 2.06e–11n = atypical 67,


mesenchymal 74

Subtype

Basa

l

Clas

sical

Mes

ench

ymal

(o)



for predicting the prognosis of HNSCC [9, 10, 37]. Accumu-lating evidence suggests that machine learning is a reliableand robust technique that can be used to quickly and accu-rately identify critical biomarkers [38, 39]. ICIs may exertbeneficial therapeutic effects in HNSCC, although theresponse rate to ICIs remains poor [15]. A lncRNA signatureof tumor-infiltrating B lymphocytes was found to havepotential applications in prognostic prediction and immu-notherapy for bladder cancer based on computational recog-nition [40], and lncRNAs associated with tumor immuneinfiltration were identified as improving clinical outcomesand immunotherapy response in non-small-cell lung cancerbased on comprehensive analysis [41]. Combinations ofceRNA with ICIs and TIICs for improving prognosis andimmunotherapy response in HNSCC are limited. Therefore,it is crucial to establish predictive biomarkers for selectingpatients who will be responsive to ICI therapy.

In the present study, we defined nine key ceRNAs(ANLN, CFL2, ITGA5, KDELC1, KIF23, NFIA, PTX3,RELT, and TMC7) and three immune cell subtypes (naïveB cells, Tregs, and neutrophils) using univariate Coxanalysis, LASSO, multivariate Cox analysis, and independentprognostic analysis. Specifically, KDELC1, PTX3, andITGA5 were found to be associated with poor prognosis inboth the K-M survival analysis and the multivariate Coxanalysis. ROC curve analysis was used to further evaluateboth signatures and confirm their favorable predictive andprognostic abilities.

We identified a potential regulatory mechanism involv-ing KCNQ1OT1 (lncRNA), miR-338-3p (miRNA)/miR-29c-3p, ITGA5 (mRNA)/KDELC1, and naïve B cells/Tregs.

A previous report indicated that KCNQ1OT1 was asso-ciated with cell proliferation, apoptosis, prognosis, invasion,and metastasis in various cancers [10, 42–44]. Further,KCNQ1OT1 facilitated invasion and inhibited apoptosis inoral squamous cell carcinoma by regulating the miR-185-5p/Rab14 axis [45]. miR-338-3p was downregulated inesophageal squamous cell carcinoma and could act as atumor suppressor [46]. In addition, miR-338-3p was shownto inhibit colorectal carcinoma cell invasion and migrationby targeting smoothened [47] and was associated with favor-able prognosis in urothelial carcinoma of the bladder [48].Inhibition of miR-29c-3p was found to be associated withpoor prognosis in patients with laryngeal squamous cell car-cinoma [49]. ITGA5 has essential roles in tumorigenesis,migration, and invasion in various cancer types [10]. Studieshave shown that ITGA5 is highly expressed in HNSCC,where its high expression is significantly associated withpoor survival [50]. Overexpression of PTX3 has been relatedto poor prognosis in pancreatic cancer and is linked to moreadvanced stages of the disease [51]. Evidence also suggeststhat PTX3 promotes metastasis of cervical cancer andEGF-induced metastasis of HNSCC through upregulationof MMP-2 and MMP-9 [52]. Our results are in accordancewith these findings and suggest that the KCNQ1OT1/miR-338-3p/ITGA5 and KCNQ1OT1/miR-29c-3p/KDELC1 axes

4

6

8Ex

pres

sion

(log2

CPM

)

2

Atyp

ical

HNSC :: RELT_expPv = 1.39e–04n = atypical 67,


mesenchymal 74

Subtype

Basa

l

Clas

sical

Mes

ench

ymal

(p)

0.0

2.5

5.0

Expr

essio

n (lo

g2CP

M)

–2.5

Atyp

ical

HNSC :: TMC7_expPv = 6.62e–09n = atypical 67,


mesenchymal 74

Subtype

Basa

l

Clas

sical

Mes

ench

ymal

(q)

Figure 6: Clinical correlation analysis. (a, b) ANLN and KIF23 expression was associated with tumor grade according to TISIDB datasets.(a) ANLN and (b) KIF23. (c–f) Key mRNAs among ceRNAs related to tumor stage according to TISIDB. (c) ITGA5, (d) KDELC1, (e) NFIA,and (f) PTX3. (g–i) Distribution of the expression of key ceRNAs across immune subtypes according to TISIDB: (g) ANLN, (h) KDELC1, (i)KIF23, and (j) NFIA. (k–q) Distribution of key ceRNA expression across molecular subtypes (TISIDB): (k) CFL2, (l) ITGA5, (m) KDELC1,(n) NFIA, (o) PTX3, (p) RELT, and (q) TMC7. Statistical significance of differential expression evaluated using Kruskal–Wallis test. C1(wound healing), C2 (IFN-gamma dominant), C3 (inflammatory), C4 (lymphocyte depleted), and C6 (TGF-β dominant). Molecularsubtypes include four types, namely, atypical (AT), basal (BA), classical (CL), and mesenchymal (MS) based on biological characteristicsof genes highly expressed in each subtype.


⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎

0.0

2.5

5.0

7.5

10.0

CD27

4

CTLA

4

HAV

CR2

LAG

3

PDCD

1

PDCD

1LG

2

TIG

IT

SIG

LEC1

5

Imm

une c

heck

poin

t

Group

Tumor

Normal

(a)

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎

⁎⁎

⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎

⁎⁎

⁎⁎

⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

⁎⁎

CD274

CTLA4

HAVCR2

LAG3

PDCD1

SIGLEC15

TIGIT

AN

LN

CFL2

ITG

A5

KIF2

3

NFI

A

PTX3

RELT

TMC7

−0.2

−0.1

0.0

0.1

0.2

0.3

⁎ p < 0.05⁎⁎ p < 0.01

Correlation

(b)



Custo

m

Rand

om

0.67

0.77

0.48

0.45

0.69 0.70

0.58

0.48

0.65

0.50

0.65

0.47

0.54

0.63

TID

E

MSI

.Sco

re

TMB

CD27

4

CD8

IFN

G

T.Cl

onal

ity

B.Cl

onal

ity

Mer

ck18

0.9

0.8

0.7

0.6

0.5

0.4 AUC

0.3

0.2

0.1

0

Uppaluri2020_PD1_HNSC_PrePos = 8, Neg = 15Uppaluri2020_PD1_HNSC_PostPos = 9, Neg = 13

(c)

RELT

ITGA5

KIF23

NFIA

ANLN

PTX3

TMC7

CFL2

BHG

712

I-BE

T-76

2JW

-7-2

4-1

KIN

001-

236

MPS

-1-I

N-1

PHA-

7938

87PI

-103

XMD

14-9

95-

Fluo

rour

acil

AR-

42BX

-912

Belin

osta

tCX

-546

1Isp

ines

ib m

esyl

ate

KIN

001-

260

Mas

itini

bM

etho

trexa

teO

SI-0

27O

SI-9

30PA

C-1

THZ-

2-49

Tuba

statin

AU

NC0

638

VX-

11e

WZ3

105

XMD

13-2

AS6

0524

0AT

-751

9A

ZD80

55BA

Y 61

-360

6BM

S345

541

CAL-

101

CAY1

0603

CP46

6722

GSK

1070

916

GSK

6906

93G

enen

tech

Cpd

10

IPA-

3KI

N00

1-10

2LA

Q82

4N

G-2

5N

PK76

-II-

72-1

Nav

itocla

xN

utlin

-3a (

-)PI

K-93

QL-

X-13

8SN

X-21

12TG

1013

48TH

Z-2-

102-

1TL

-1-8

5TL

-2-1

05TP

CA-1

VN

LG/1

24Vo

rinos

tat

YM20

1636

ZSTK

474

AIC

AR

AP-

2453

4BI

X021

89CU

DC-

101

GSK

2126

458

KIN

001-

244

Phen

form

inQ

L-XI

-92

QL-

XII-

47SB

5233

4TA

K-71

5AT

RAFK

866

Ruxo

litin

ibA

fatin

ibCC

T007

093

Erlo

tinib

6816

40A-

7700

41Bi

calu

tam

ide

Borte

zom

ibCG

P-08

2996

Cetu

xim

abD

asat

inib

GSK

1904

529A

Gefi

tinib

Lapa

tinib

PF-5

6227

1W

Z-1-

84BI

RB 0

796

Bleo

myc

in (5

0 uM

)CH

IR-9

9021

IOX2

JNK

Inhi

bito

r VIII

LFM

-A13

Sara

catin

ibW

H-4

-023

XAV

939

Mid

osta

urin

SB 2

1676

3Pi

perlo

ngum

ine

17-A

AGEl

esclo

mol

RO-3

306

TGX2

21D

ocet

axel

Drug

Sym

bol

FDR< 0.05> 0.05

0.4

0.0

−0.4CorrelationCorrelation between GDSC drug sensitivity and mRNA expression

(d)

Figure 7: Key ceRNA expression correlates with immunotherapy response and drug sensitivity. (a) Differences in expression of tumor-related immunosuppressive molecules in HNSCC; ∗P < 0:05, ∗∗P < 0:01, and ∗∗∗P < 0:001. (b) Correlations between key ceRNAs and ICmolecules. The horizontal and vertical coordinates represent genes. Different colors represent correlation coefficients (blue, positivecorrelation and red, negative correlation); the darker the color, the stronger the correlation; ∗P < 0:05 and ∗∗P < 0:01. (c) Comparison ofAUC values between the custom biomarker and other published biomarkers in predicting anti-PD1 response. (d) Correlation analysis ofdrug sensitivity and expression of key ceRNAs.

Table 2: Survival analysis for major target ceRNAs using LOGpc.

Symbol Dataset HR 95% CI P value Prognostic

CFL2 TCGA 1.3608 1.0137-1.8268 0.0403 Poor

ITGA5 GSE31056 2.6804 1.3752-5.2245 0.0038 Poor

ITGA5 TCGA 1.4878 1.1162-1.9832 0.0067 Poor

KDELC1 TCGA 1.4946 1.1212-1.9922 0.0061 Poor

KDELC1 GSE31056 2.1694 1.0960-4.2940 0.0262 Poor

NFIA GSE65858 0.4627 0.2648-0.8086 0.0068 Good

PTX3 GSE31056 3.9765 2.0435-7.7379 <0.0001 Poor

PTX3 TCGA 1.3605 1.0103-1.8320 0.0426 Poor

RELT TCGA 1.3554 1.0113-1.8165 0.0418 Poor


Decreased risk (P < 0.05, Z > 0)Increased risk (P < 0.05, Z > 0)Not significant (P > 0.05)

HNSC (n = 522) ANLN

HNSC-HPV+ (n = 98) TMC7

Cancer Gene P z_score0.0520 1.9431

0.0000 4.2508

HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)HNSC (n = 522)HNSC-HPV– (n = 422)

ANLNANLNCFL2CFL2CFL2

ITGA5ITGA5ITGA5

KDELC1KDELC1KDELC1

KIF23KIF23KIF23NFIANFIANFIAPTX3PTX3PTX3RELTRELTRELTTMC7TMC7

0.04720.42890.01000.01140.31960.00030.05290.00010.00210.03000.00670.15380.05080.87770.00060.08150.00010.00090.00270.16220.09490.03260.38730.04330.7202

1.98440.79102.57452.53130.99523.61161.93554.02233.07992.17032.71111.42611.95310.1539

–3.4135–1.7420–3.81623.33593.00461.39771.66992.1370

–0.86452.02100.3583

(a)

p > 0.05

p≤ 0.05

HNSC (n = 522)Zscore

4.7

0.0

–4.1

B ce

ll_TI

MER

HNSC-HPV– (n = 422)HNSC-HPV+ (n = 98)

B ce

ll_EP

ICB

cell_

QUA

NTI

SEQ

B ce

ll_XC

ELL

B ce

ll_M

CPCO

UN

TER

B ce

ll m

emor

y_CI

BERS

ORT

B ce

ll m

emor

y_CI

BERS

ORT

-ABS

B ce

ll m

emor

y_XC

ELL

B ce

ll na

ive_

CIBE

RSO

RTB

cell

naiv

e_CI

BERS

ORT

-ABS

B ce

ll na

ive_

XCEL

LB

cell

plas

ma_

CIBE

RSO

RTB

cell

plas

ma_

CIBE

RSO

RT-A

BSB

cell

plas

ma_

XCEL

LCl

ass-

switc

hed

mem

ory

B ce

ll_XC

ELL

(b)



p > 0.05

p ≤ 0.05

HNSC (n = 522)

Zscore4.7

0.0

–4.1

Neu

troph

il_TI

MER

HNSC-HPV– (n = 422)

HNSC-HPV+ (n = 98)

Neu

troph

il_CI

BERS

ORT

Neu

troph

il_CI

BERS

ORT

-ABS

Neu

troph

il_M

CPCO

UN

TER

Neu

troph

il_Q

UAN

TISE

Q

Neu

troph

il_XC

ELL

(c)

p > 0.05

p≤ 0.05

HNSC (n = 522)Zscore

4.7

0.0

–4.1

T ce

ll re

gulat

ory

(treg

s)_C

IBER

SORT

HNSC-HPV– (n = 422)

HNSC-HPV+ (n = 98)

T ce

ll re

gulat

ory

(treg

s)_C

IBER

SORT

-ABS

T ce

ll re

gulat

ory

(treg

s)_Q

UAN

TISE

Q

T ce

ll re

gulat

ory

(treg

s)_X

CELL

(d)

CD274

CTLA4

HAVCR2

LAG3

PDCD1

PDCD1LG2

SIGLEC15

TIGIT

HNSC.normalHNSC-HPVpos.tumorHNSC-HPVneg.tumor

HNSC.tumor

HNSC.tumor

HNSC.normalHNSC-HPVpos.tumorHNSC-HPVneg.tumor

HNSC.tumorHNSC.normal

HNSC-HPVpos.tumorHNSC-HPVneg.tumor











⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎⁎

⁎⁎

(e)



have important roles in determining the prognosis ofHNSCC patients.

The associations among the nine key ceRNA signatures,three immune cell signatures, and different clinicopathologi-cal features was assessed. The results showed that ANLN,KIF23, and ITGA5 expression was significantly correlatedwith grade. Increased KDELC1 and PTX3 expression wasrelated to advanced stage, whereas overexpression of NFIAwas correlated with low tumor stage (Figures 6(a)–6(f)). Arecent study showed that the C1 and C2 subtypes were

enriched in HNSCC and were associated with less favorableoutcomes. The C3 subtype had the most favorable prognosis,whereas the C4 and C6 subtypes had the worst prognoses[53]. In our study, we observed increased expression ofANLN and KIF23 in subtypes C1 and C2, whereas KDELC1showed higher expression in the C1, C2, and C6 subtypes,indicating that it may be associated with worse survival.NFIA showed increased expression in subtype C3 anddecreased expression in C4, which indicated that it couldpredict better survival (Figures 6(g)–6(j)).

(f)

Figure 8: Validation of the prognostic value of identified biomarkers and associations between key ceRNAs and IC genes. (a) Confirmationof prognostic value of key ceRNAs in HNSCC with different HPV statuses via the TIMER 2.0 database. (b–d) Analysis of associationsbetween key immune cells and OS in HNSCC patients with different HPV statuses based on Cox regression analysis using the TIMER2.0 database (http://timer.cistrome.org/). (e) Differences in expression of IC genes with different HPV statuses in HNSCC using TIMER2.0. (f) Exploration of the associations between key ceRNAs and IC molecules with different HPV statuses in HNSCC using TIMER 2.0.



Recently, Walter et al. [13] reported that the MS subtypewas associated with distant metastases, and that treatmentwith EGFR inhibitors was less likely to be effective in theAT subtype. Our results indicated that CFL2, ITGA5,KDELC1, and PTX3 were highly expressed in the MS sub-type. Higher NFIA expression was associated with the ATsubtype (Figures 6(k)–6(q)). These studies further supportour finding that NFIA expression was correlated with favor-able outcomes whereas higher expression of the other ceR-NAs was associated with unfavorable outcomes.

Previous studies have revealed that naïve B cells andTregs are indicators of better survival, whereas neutrophilshave been associated with HPV positivity and pooreroutcomes [7, 10, 54]. Higher Treg infiltration was found tobe correlated with better prognosis and longer survival inHPV-positive patients compared with HPV-negativeHNSCC patients [55]. By contrast, naïve B cells could becorrelated with tumorigenesis and progression of HNSCC[10]. Neutrophils induce tumor progression and promotetumor cell migration in HNSCC; therefore, neutrophil infil-tration represents a risk factor. [56, 57]. Our integrated anal-yses showed that three immune cell subtypes (naive B cells,Tregs, and neutrophils) could be used as prognosis-relatedbiomarkers in HNSCC. Naïve B cells and Tregs were associ-ated with favorable prognosis, whereas neutrophils repre-sented an unfavorable prognostic marker in HNSCC,especially in HPV-positive HNSCC patients (Figures 4(e)and 8(b)–8(d)). ROC analysis showed that the key immunecell-related model constructed in this study could predictthe prognosis of patients with HNSCC. As shown inFigures 5(j)–5(l), higher levels of naïve B cells and Tregswere related to lower risk scores (P = 0:0085 and P = 4e − 9). These results further suggested that these immune celltypes (naïve B cells, Tregs, and neutrophils) were related toprognosis. Our results are consistent with those of previousstudies, and our immune cell-related model could be usedas an independent factor to assess prognosis in HNSCC.

Finally, we found that eight IC genes exhibited higherexpression in tumor tissues and were correlated with key ceR-

NAs in HNSCC with different HPV statuses (Figures 7(a)and 7(b) and 8(e) and 8(f)). Our identified biomarker couldmore effectively predict anti-PD1 response compared withpublished biomarkers (Figure 7(c)). Previous studies havereported that HPV-positive HNCs expressed high levels ofmultiple T cell exhaustion markers, including LAG3, PD1,TIGIT, and TIM3; the high expression of these markers wascorrelated with improved survival in HPV-positive HNCpatients [58, 59]. Based on these studies and our findings,HNSCC may exhibit strong beneficial responses to immuno-therapy and high expression levels of these key genes maycontribute to predicting improved survival of HNSCCpatients, especially in cases of HPV-positive HNSCC. We alsoobserved that eight key genes were associated with treatmentresponse to most drugs, suggesting the potential of these mol-ecules as therapeutic drug targets in HNSCC (Figure 7(d)).These results provide a scientific rationale for potentiallycombining ICI therapy with inhibitors of the eight key genesidentified in this study to improve treatment efficacy forHNSCC patients.

5. Conclusions

We identified nine key ceRNAs and three immune cell-related signatures as potential biomarkers for predictingthe prognosis of HNSCC. In addition, the KCNQ1OT1(lncRNA), miR-338-3p (miRNA)/miR-29c-3p, ITGA5(mRNA)/KDELC1, and naïve B cell/Treg axes may be linkedto prognosis of HNSCC. As key IC-related members ofceRNAs may be associated with response to ICI therapy,combining ICI with inhibitors of these eight key genesmay contribute to improving treatment efficacy in HNSCpatients. However, additional clinical data and experimentsare required to confirm the prognostic value of thesesignatures and their potential associations with ICI immu-notherapy outcomes in patients with HNSCC. We willperform further experiments to confirm the currentresearch results.

Table 3: Associations between gene expression and therapy outcome in clinical studies of IC blockade.

Gene Cohort Cancer Subtype Survival Risk.adj

ANLNRiaz2017_PD1 Melanoma Ipi_Prog OS 2.025

Riaz2017_PD1 Melanoma Ipi_naive PFS -3.136

CFL2 Liu2019_PD1 Melanoma Ipi_naive OS 2.489

ITGA5Mariathasan2018_PDL1 Bladder mUC OS 3.52

Miao2018_ICB Kidney Clear PFS -3.173

KIF23 Riaz2017_PD1 Melanoma Ipi_Prog OS 2.802

NFIAMariathasan2018_PDL1 Bladder mUC OS 2.987

Zhao2019_PD1 Glioblastoma Pre PFS -2.336

PTX3 Riaz2017_PD1 Melanoma Ipi_naive OS 2.467

RELTMariathasan2018_PDL1 Bladder mUC OS 2.505

Liu2019_PD1 Melanoma Ipi_naive PFS -2.32

TMC7Hugo2016_PD1 Melanoma OS 2.27

Lauss2017_ACT Melanoma OS -2.528


Data Availability

The gene expression RNA-sequencing data and clinicalinformation of HNSCC in the current study were obtainedfrom TCGA data portal (https://portal.gdc.cancer.gov/).

Conflicts of Interest

The authors declare that there is no conflict of interestregarding the publication of this manuscript.

Authors’ Contributions

Ya Guo conceived and designed the article, performed dataanalysis, and wrote the main text of the article. Wei KangPan reviewed the manuscript. Zhong Wei Wang performedlanguage editing. Wang Hui Su and Kun Xu preparedFigures 1–4. Hui Jia and Jing Chen prepared Figures 5–8and tables. Ya Guo and Wei Kang Pan contributed equallyto this work and are co-first authors.

Acknowledgments

This work was supported by the Shaanxi Provincial Natu-ral Science Foundation (grant number 2017JQ8057). Wethank Zhong Guo Liang for bioinformatics-related techni-cal support.

Supplementary Materials

Supplementary 1. Additional file 1. Original source data:online web service.

Supplementary 2. Additional file 2. Original source data:related R software.

References

[1] L. Zhang, B. Li, Y. Peng et al., “The prognostic value of TMBand the relationship between TMB and immune infiltrationin head and neck squamous cell carcinoma: a geneexpression-based study,” Oral Oncology, vol. 110, article104943, 2020.

[2] Y. Zhang, A. Lin, Y. Li et al., “Age and mutations as predictorsof the response to immunotherapy in head and neck squamouscell cancer,” Frontiers in cell and developmental biology, vol. 8,article 608969, 2020.

[3] J. Li, R. M. Srivastava, A. Ettyreddy, and R. L. Ferris, “Cetuxi-mab ameliorates suppressive phenotypes of myeloid antigenpresenting cells in head and neck cancer patients,” Journalfor Immunotherapy of Cancer, vol. 3, no. 1, p. 54, 2015.

[4] E. E. W. Cohen, R. B. Bell, C. B. Bifulco et al., “The Society forImmunotherapy of Cancer consensus statement on immuno-therapy for the treatment of squamous cell carcinoma of thehead and neck (HNSCC),” Journal for Immunotherapy of Can-cer, vol. 7, no. 1, p. 184, 2019.

[5] Y. Pan, G. Liu, D. Wang, and Y. Li, “Analysis of lncRNA-mediated ceRNA crosstalk and identification of prognostic sig-nature in head and neck squamous cell carcinoma,” Frontiersin Pharmacology, vol. 10, p. 150, 2019.

[6] A. Jiang, N. Liu, S. Bai et al., “The construction and analysisof tumor-infiltrating immune cells and ceRNA networks in

bladder cancer,” Frontiers in Genetics, vol. 11, article605767, 2020.

[7] J. Song, Z. Deng, J. Su, D. Yuan, J. Liu, and J. Zhu, “Patterns ofimmune infiltration in HNC and their clinical implications: agene expression-based study,” Frontiers in Oncology, vol. 9,p. 1285, 2019.

[8] X. Zhang, M. Shi, T. Chen, and B. Zhang, “Characterization ofthe immune cell infiltration landscape in head and neck squa-mous cell carcinoma to aid immunotherapy,” Molecular ther-apy Nucleic acids, vol. 22, pp. 298–309, 2020.

[9] X.-N. Fang, M. Yin, H. Li et al., “Comprehensive analysis ofcompetitive endogenous RNAs network associated with headand neck squamous cell carcinoma,” Scientific Reports, vol. 8,no. 1, article 10544, 2018.

[10] H. Zhou, Y. He, L. Li, C. Wu, and G. Hu, “Identification novelprognostic signatures for head and neck squamous cell carci-noma based on ceRNA network construction and immuneinfiltration analysis,” International Journal of Medical Sciences,vol. 18, no. 5, pp. 1297–1311, 2021.

[11] R. L. Ferris, “Immunology and immunotherapy of head andneck cancer,” Journal of Clinical Oncology: Official Journal ofthe American Society of Clinical Oncology, vol. 33, no. 29,pp. 3293–3304, 2015.

[12] X. Xu, J. Li, J. Zou et al., “Association of germline variants innatural killer cells with tumor immune microenvironmentsubtypes, tumor-infiltrating lymphocytes, immunotherapyresponse, clinical outcomes, and cancer risk,” JAMA networkopen, vol. 2, no. 9, article e199292, 2019.

[13] V. Walter, X. Yin, M. D. Wilkerson et al., “Molecular subtypesin head and neck cancer exhibit distinct patterns of chromo-somal gain and loss of canonical cancer genes,” PloS one,vol. 8, no. 2, article e56823, 2013.

[14] S. Farkona, E. P. Diamandis, and I. M. Blasutig, “Cancerimmunotherapy: the beginning of the end of cancer?,” BMCMedicine, vol. 14, no. 1, p. 73, 2016.

[15] L. Yi, G. Wu, L. Guo, X. Zou, and P. Huang, “Comprehensiveanalysis of the PD-L1 and immune infiltrates of mA RNAmethylation regulators in head and neck squamous cell carci-noma,” Molecular therapy Nucleic acids, vol. 21, pp. 299–314,2020.

[16] G. Wang, D. Xu, Z. Zhang et al., “The pan-cancer landscape ofcrosstalk between epithelial-mesenchymal transition andimmune evasion relevant to prognosis and immunotherapyresponse,” NPJ precision oncology, vol. 5, no. 1, p. 56, 2021.

[17] B. Feng and J. Hess, “Immune-related mutational landscapeand gene signatures: prognostic value and therapeutic impactfor head and neck cancer,” Cancers, vol. 13, no. 5, p. 1162,2021.

[18] Y. Yao, Z. Yan, S. Lian et al., “Prognostic value of novelimmune-related genomic biomarkers identified in head andneck squamous cell carcinoma,” Journal for immunotherapyof cancer, vol. 8, no. 2, article e000444, 2020.

[19] C. Ili, K. Buchegger, H. Demond et al., “Landscape of genome-wide DNA methylation of colorectal cancer metastasis,” Can-cers, vol. 12, no. 9, p. 2710, 2020.

[20] E. Pérez-Riesgo, L. Gutiérrez, D. Ubierna et al., “Transcripto-mic analysis of calcium remodeling in colorectal cancer,”International journal of molecular sciences, vol. 18, no. 5,p. 922, 2017.

[21] Y. Zhu, Y. Bian, Q. Zhang et al., “Construction and analysis ofdysregulated lncRNA- associated ceRNA network in colorectal


https://portal.gdc.cancer.gov/

https://downloads.hindawi.com/journals/bmri/2021/4532438.f1.docx

https://downloads.hindawi.com/journals/bmri/2021/4532438.f2.docx

cancer,” Journal of Cellular Biochemistry, vol. 120, no. 6,pp. 9250–9263, 2019.

[22] P. Shannon, A. Markiel, O. Ozier et al., “Cytoscape: a softwareenvironment for integrated models of biomolecular interac-tion networks,” Genome Research, vol. 13, no. 11, pp. 2498–2504, 2003.

[23] R. Huang, J. Wu, Z. Zheng et al., “The construction and anal-ysis of ceRNA network and patterns of immune infiltration inmesothelioma with bone metastasis,” Frontiers in Bioengineer-ing and Biotechnology, vol. 7, p. 257, 2019.

[24] K. Lin, L.-J. Song, J. Ma, T.-S. Zhang, D.-Y. You, and Y.-W. He,“Identification of cancer hallmark-associated gene andlncRNA cooperative regulation pairs and dictate lncRNA rolesin oral squamous cell carcinoma,” Journal of Cellular andMolecular Medicine, vol. 24, no. 9, pp. 5213–5223, 2020.

[25] Y. Guo, P. T. Yang, Z. W. Wang, K. Xu, W. H. Kou, andH. Luo, “Identification of three autophagy-related long non-coding RNAs as a novel head and neck squamous cell carci-noma prognostic signature,” Frontiers in Oncology, vol. 10,article 603864, 2021.

[26] S. Liu, A. Song, X. Zhou et al., “ceRNA network developmentand tumour-infiltrating immune cell analysis of metastaticbreast cancer to bone,” Journal of bone oncology, vol. 24, article100304, 2020.

[27] J. Lin, M. Yu, X. Xu et al., “Identification of biomarkers relatedto CD8 T cell infiltration with gene co-expression network inclear cell renal cell carcinoma,” Aging, vol. 12, no. 4,pp. 3694–3712, 2020.

[28] P. Jiang, S. Gu, D. Pan et al., “Signatures of T cell dysfunctionand exclusion predict cancer immunotherapy response,”Nature Medicine, vol. 24, no. 10, pp. 1550–1558, 2018.

[29] C.-J. Liu, F.-F. Hu, M. Xia, L. Han, Q. Zhang, and A. Guo,“GSCALite: a web server for gene set cancer analysis,” Bioinfor-matics (Oxford, England), vol. 34, no. 21, pp. 3771-3772, 2018.

[30] W. Yang, J. Soares, P. Greninger et al., “Genomics of Drug Sen-sitivity in Cancer (GDSC): a resource for therapeutic bio-marker discovery in cancer cells,” Nucleic Acids Research,vol. 41, no. Database issue, pp. D955–D961, 2013.

[31] G. Wu, Y. Xu, N. Ruan et al., “Genetic alteration and clinicalsignificance of SUMOylation regulators in multiple cancertypes,” Journal of Cancer, vol. 11, no. 23, pp. 6823–6833, 2020.

[32] H. Zheng, G. Zhang, L. Zhang et al., “Comprehensive review ofweb servers and bioinformatics tools for cancer prognosisanalysis,” Frontiers in Oncology, vol. 10, p. 68, 2020.

[33] B. Hu, X.-B. Yang, and X.-T. Sang, “Development of animmune-related prognostic index associated with hepatocel-lular carcinoma,” Aging, vol. 12, no. 6, pp. 5010–5030, 2020.

[34] T. Li, J. Fu, Z. Zeng et al., “TIMER2.0 for analysis of tumor-infiltrating immune cells,” Nucleic Acids Research, vol. 48,no. W1, pp. W509–W514, 2020.

[35] C. R. Leemans, B. J. M. Braakhuis, and R. H. Brakenhoff, “Themolecular biology of head and neck cancer,” Nature ReviewsCancer, vol. 11, no. 1, pp. 9–22, 2011.

[36] Z. X. Li, Z. Q. Zheng, Z. H. Wei et al., “Comprehensive charac-terization of the alternative splicing landscape in head andneck squamous cell carcinoma reveals novel events associatedwith tumorigenesis and the immune microenvironment,” The-ranostics, vol. 9, no. 25, pp. 7648–7665, 2019.

[37] A.-M. Jiang, M.-D. Ren, N. Liu et al., “Tumor mutation bur-den, immune cell infiltration, and construction of immune-related genes prognostic model in head and neck cancer,”

International Journal of Medical Sciences, vol. 18, no. 1,pp. 226–238, 2021.

[38] Z. Wang, H. Yuan, J. Huang et al., “Prognostic value ofimmune-related genes and immune cell infiltration analysisin the tumor microenvironment of head and neck squamouscell carcinoma,” Head & Neck, vol. 43, no. 1, pp. 182–197,2021.

[39] Y. Chen, Y. Li, R. Narayan, A. Subramanian, and X. Xie, “Geneexpression inference with deep learning,” Bioinformatics(Oxford, England), vol. 32, no. 12, pp. 1832–1839, 2016.

[40] M. Zhou, Z. Zhang, S. Bao et al., “Computational recognitionof lncRNA signature of tumor-infiltrating B lymphocytes withpotential implications in prognosis and immunotherapy ofbladder cancer,” Briefings in Bioinformatics, vol. 22, no. 3,2021.

[41] J. Sun, Z. Zhang, S. Bao et al., “Identification of tumor immuneinfiltration-associated lncRNAs for improving prognosis andimmunotherapy response of patients with non-small cell lungcancer,” Journal for immunotherapy of cancer, vol. 8, no. 1,article e000110, 2020.

[42] Y. Zhu, Y. Shen, R. Chen et al., “KCNQ1OT1 lncRNA affectsthe proliferation, apoptosis, and chemoresistance of small celllung cancer cells via the JAK2/STAT3 axis,” Annals of transla-tional medicine, vol. 9, no. 10, p. 891, 2021.

[43] H. Hao, H. Chen, L. Xie, H. Liu, and D. Wang, “LncRNAKCNQ1OT1 promotes proliferation, invasion and metastasisof prostate cancer by regulating miR-211-5p/CHI3L1 path-way,” Oncotargets and Therapy, vol. 14, pp. 1659–1671, 2021.

[44] X. Zhong, X. Wen, L. Chen, N. Gu, X. Yu, and K. Sui, “Longnon-coding RNA KCNQ1OT1 promotes the progression ofgastric cancer via the miR-145-5p/ARF6 axis,” The Journal ofGene Medicine, vol. 23, no. 5, article e3330, 2021.

[45] Q. Bao, X. Liao, R. Li, and N. Ding, “KCNQ1OT1 promotesmigration and inhibits apoptosis by modulating miR-185-5p/Rab14 axis in oral squamous cell carcinoma,” Development,Growth & Differentiation, vol. 61, no. 9, pp. 466–474, 2019.

[46] M. Yang, R. Liu, J. Sheng et al., “Differential expression profilesof microRNAs as potential biomarkers for the early diagnosisof esophageal squamous cell carcinoma,” Oncology Reports,vol. 29, no. 1, pp. 169–176, 2013.

[47] Q. Xue, K. Sun, H.-J. Deng, S.-T. Lei, J.-Q. Dong, and G.-X. Li,“MicroRNA-338-3p inhibits colorectal carcinoma cell inva-sion and migration by targeting smoothened,” Japanese Jour-nal of Clinical Oncology, vol. 44, no. 1, pp. 13–21, 2014.

[48] T. Inamoto, H. Uehara, Y. Akao et al., “A panel of microRNAsignature as a tool for predicting survival of patients withurothelial carcinoma of the bladder,” Disease Markers,vol. 2018, Article ID 5468672, 6 pages, 2018.

[49] R. Fang, Y. Huang, J. Xie, J. Zhang, and X. Ji, “Downregulationof miR-29c-3p is associated with a poor prognosis in patientswith laryngeal squamous cell carcinoma,” Diagnostic Pathol-ogy, vol. 14, no. 1, p. 109, 2019.

[50] S. Kisoda, W. Shao, N. Fujiwara et al., “Prognostic value of par-tial EMT-related genes in head and neck squamous cell carci-noma by a bioinformatic analysis,”Oral Diseases, vol. 26, no. 6,pp. 1149–1156, 2020.

[51] S. Kondo, H. Ueno, H. Hosoi et al., “Clinical impact of pentra-xin family expression on prognosis of pancreatic carcinoma,”British Journal of Cancer, vol. 109, no. 3, pp. 739–746, 2013.

[52] S.-H. Chan, J. P. Tsai, C.-J. Shen, Y.-H. Liao, and B. K. Chen,“Oleate-induced PTX3 promotes head and neck squamous cell


carcinoma metastasis through the up-regulation of vimentin,”Oncotarget, vol. 8, no. 25, pp. 41364–41378, 2017.

[53] V. Thorsson, D. L. Gibbs, S. D. Brown et al., “The immunelandscape of cancer,” Immunity, vol. 48, no. 4, pp. 812–830.e14, 2018.

[54] J. Zhang, X. Zhong, H. Jiang et al., “Comprehensive character-ization of the tumor microenvironment for assessing immuno-therapy outcome in patients with head and neck squamous cellcarcinoma,” Aging, vol. 12, no. 22, pp. 22509–22526, 2020.

[55] X. Liu, J. Chen, W. Lu et al., “Systematic profiling of immunerisk model to predict survival and immunotherapy responsein head and neck squamous cell carcinoma,” Frontiers inGenetics, vol. 11, article 576566, 2020.

[56] N. Borregaard, “Neutrophils, from marrow to microbes,”Immunity, vol. 33, no. 5, pp. 657–670, 2010.

[57] C. A. Dumitru, H. Gholaman, S. Trellakis et al., “Tumor-derived macrophage migration inhibitory factor modulatesthe biology of head and neck cancer cells via neutrophil activa-tion,” International Journal of Cancer, vol. 129, no. 4, pp. 859–869, 2011.

[58] S. F. Gameiro, F. Ghasemi, J. W. Barrett et al., “Treatment-naïve HPV+ head and neck cancers display a T-cell-inflamedphenotype distinct from their HPV- counterparts that hasimplications for immunotherapy,” Oncoimmunology, vol. 7,no. 10, article e1498439, 2018.

[59] W. Qiang, Y. Dai, X. Xing, and X. Sun, “Identification and val-idation of a prognostic signature and combination drug ther-apy for immunotherapy of head and neck squamous cellcarcinoma,” Computational and Structural BiotechnologyJournal, vol. 19, pp. 1263–1276, 2021.


identification of novel biomarkers for predicting

Documents