MEDICAL SCIENCE MONITOR

Received:2020.09.13
Accepted:2020.12.15
Available online:2020.12.31
Published:2021.03.05

Identification of NDRG Family Member 4 (NDRG4) and CDC28 Protein Kinase Regulatory Subunit 2 (CKS2) as Key Prognostic Genes in Adrenocortical Carcinoma by Transcriptomic Analysis

Department of Urology, Shanghai Yangsi Hospital, Shanghai, P.R. China

Authors’ Contribution: Study Design A Data Collection B

Statistical Analysis C Data Interpretation D Manuscript Preparation E Literature Search F Funds Collection G

BE Zhengqing Yang

CD Hui Cheng

F

Yazhou Zhang

AG Yan Zhou

Corresponding Author: Source of support:

Yan Zhou, e-mail: zhou.yanys@hotmail.com Departmental sources

Background: Adrenocortical carcinoma (ACC) is an aggressive cancer with heterogeneous outcomes. In this study, we aimed to investigate genomic and prognostic features of ACC.

Material/Methods: Clinical, pathologic, and transcriptomic data from 2 independent datasets derived from ACC samples (TCGA- ACC dataset, GEO-GSE76021 dataset) were collected. Weighted gene co-expression network analysis (WGCNA) and survival analyses were performed to identify prognostic genes. Pathway analysis was performed for mech- anistic analysis. xCell deconvolution was performed for tumor microenvironment analysis.

Results: In the TCGA-ACC cohort, WGCNA identified a prognostic module of 5408 genes. Differential expression analysis identified 1969 genes that differed in expression level between long-term and short-term survivors. Univariate Cox regression model analysis identified 8393 genes with prognostic value. The intersection of these gene sets included 820 prognostic genes. Similar protocols were performed for the GSE76021 dataset, and 5 candidate genes were identified. Further intersection of these genes finally identified NDRG4 and CKS2 as key prognos- tic genes. Multivariate Cox regression model analysis validated the prognostic value of NDRG4 (HR=0.61, 95% CI 0.46-0.80) and CKS2 (HR=2.52, 95% CI 1.38-4.60). Moreover, NDRG4 and CKS2 expression predicted surviv- al in patients treated with mitotane (P<0.001). Further mechanism exploration found an association between CKS2 and DNA mismatch repair pathways. Moreover, NDRG4 positively correlated with CD8+ T cell infiltration, while CKS2 negatively correlated with it.

Conclusions:

We identified NDRG4 and CKS2 expression as key prognostic genes in ACC, which may help in risk stratifica- tion of ACC. Moreover, a close relationship was found between CKS2 and mismatch repair pathways. Moreover, immune cell infiltration differed according to NDRG4 and CKS2 expression.

Keywords: Adrenocortical Carcinoma . Biological Markers . Prognosis . Transcriptome

Full-text PDF:

https://www.medscimonit.com/abstract/index/idArt/928523

3256 1 7

2] 41

Background

Adrenocortical carcinoma (ACC) is an aggressive cancer orig- inating in the cortex of the adrenal gland [1]. Almost half of ACC patients will experience recurrence or metastasis [2], re- sulting in a relatively low 5-year overall survival (OS) rate, rang- ing from 16% to 60% [3,4]. Traditional prognostic factors for ACC include tumor stage [5], resection status [6], and Ki67 in- dex of tumor proliferation [7]. However, these traditional fac- tors still have limitations. Recent studies have identified nov- el molecular biomarkers for predicting ACC survival, such as a 5-gene (TOP2A, NDC80, CEP55, CDKN3, and CDK1) signature, as well as DNA damage and cell cycle pathways [8-12], indi- cating that the molecular pathogenesis of ACC may also be involved in disease progression and survival.

Advances in molecular biology have uncovered part of the pathogenesis mechanism of ACC [13-17], but the precise mech- anisms are still far from known. Recently, genomic approach- es have provided a more comprehensive landscape for ACC pathogenesis [18,19]. Assié et al were the first to apply exome sequencing and SNP array analysis in a relatively large num- ber of ACC patients [19]. They identified 2 distinct molecular ACC subgroups with opposite outcome [19]. Another study by Zheng et al applied a comprehensive pan-genomic analysis of

ACC patients, which identified PRKAR1A, RPL22, TERF2, CCNE1, and NF1 as new driver genes of ACC [18]. They also divided ACC into 3 subtypes according to a DNA-methylation signature, which may be helpful in the risk stratification of ACC [18]. These high-throughput genomic studies helped to expand knowledge boundaries with regard to ACC development.

Mitotane (1,1-(dichlorodiphenyl)-2,2-dichloroethane) is a ste- roidogenesis inhibitor and cytostatic antineoplastic medication for ACC treatment, which is recommended as systemic therapy for metastatic or unresectable ACC, as well as adjuvant ther- apy for high-risk postoperative patients [20]. However, not all patients respond to mitotane. Recently, the pharmacokinetics of mitotane has been found to be a predictive factor in ACC prognosis [21,22]. A recent study also identified the correla- tion between SOAT1 expression and response to mitotane in ACC patients [23]. However, no robust predictive biomarker has been applied in the clinic.

In the present study, by re-evaluation of 2 open-access tran- scriptomics datasets, we aimed to identify key prognostic genes in ACC, which may be predictive of the efficacy of mi- totane therapy as well. Moreover, we attempted to discover the molecular mechanisms underlying the functions of these prognostic genes.

Figure 1. Flow chart of the study design.

12.5

Down Stable

-

10.0

Up

TCGA-ACC

-log10 (p-value)

7.5

GSE76021

5.0

.

25

0.0

WGCNA

Cox regression

DEGS

-2.5

log2 (fold change)

0.0

2.5

1.00

NDRGA level +high+low

1.00

CK52 level

-high +-low

100

Survival probabil ity

Survival probabil ity

CKS2 NDRG4

50

50

80

1.

0.25

p -< 0.001

0.25

p<0.001

60

0.00

0 1 2 3 4 5 6 7 8 9101112 Tomat (ytum)

0.00

Sensivity

01234567231711

40

1.00

NDAG4 +high+lov

1.00

CK52 +high +low

Survival probability

Survival probabil ity

D.TS

Multivariate cox

Kaplan-Meier analysis

Receiver operating characteristic analysis

20

Stage

AUC

Pvalue

0.2001

50

0.50

NORG4 Risk score

0.906

0.974

0.1182

0.0016

0.75

0

p<0.001

0.25

p<0.001

0

20

40

60

80

100

0.00

1123456713101112

0.00

8123456713112

100-specificity

Pathway analysis

4

2

0

-

Immunologic analysis

8

2

2

2

0

2

6

0

6

8

0

6

8

0

6

8

4

4

4

4

BRCA1

2

8

2

2

2

0

6

0

6

8

0

6

8

0

6

8

4

4

4

4

Table 1. Baseline characteristics of the TCGA-ACC cohort.
Male (%)Female (%)Total (%)
Sex31 (39.2%)48 (60.8%)79 (100%)
Age (years)48.74±15.6745.38+15.6946.70±15.67
Status
Dead11 (13.9%)17 (21.5%)28 (35.4%)
Alive20 (25.3%)31 (39.2%)51 (64.6%)
T stage
13 (3.8%)6 (7.6%)9 (11.4%)
215 (19.0%)27 (34.2%)42 (53.2%)
34 (5.0%)4 (5.0%)8 (10.1%)
47 (8.9%)11 (13.9%)18 (22.8%)
N stage
027 (34.2%)41 (51.9%)68 (86.1%)
12 (2.5%)7 (8.9%)9 (11.4%)
M stage
024 (30.4%)38 (48.1%)62 (78.5%)
15 (6.3%)10 (12.7%)15 (19.0%)
Stage
I3 (3.8%)6 (7.6%)9 (11.4%)
II15 (19.0%)22 (27.8%)37 (46.8%)
III6 (7.6%)10 (12.7%)16 (20.3%)
6IV5 (6.3%)10 (12.7%)15 (19.0%)
Events
Local recurrence5 (6.3%)5 (6.3%)10 (12.7%)
Distant metastasis6 (7.6%)20 (25.3%)26 (32.9%)
New primary tumor1 (1.3%)0 (0.0%)1 (1.3%)

Material and Methods

Study Populations and Data Acquisition

The study design is illustrated in Figure 1. Two independent datasets, the TCGA-ACC dataset and the GEO-GSE76021 da- taset, were downloaded for the study. The TCGA-ACC cohort included 79 patients (Table 1) and the GEO-GSE76021 co- hort included 29 patients. For the TCGA-ACC cohort, clinical, pathologic, and transcriptomic data were downloaded from the UCSC Xena (https://xena.ucsc.edu/) website. For the GEO- GSE76021 cohort, clinical, pathologic, and transcriptomic data were downloaded from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) website. R software (version 4.0.2; https://www.r-project.org/) and Bioconductor packages (http://www.bioconductor.org/) were used for data analysis.

The “sva” package [24] was used for the normalization of the GSE76021 dataset. For duplicates, RNA expression was aver- aged, and genes with relatively low abundance were discard- ed. We followed the access policies of the TCGA and GEO da- tabases strictly during the process of the study.

Weighted Gene Co-expression Network Analysis

Weighted gene co-expression network analysis (WGCNA) was used as a systematic method to describe the pattern of gene correlation, as previously described by Langfelder et al [25]. We set OS as the phenotype for WGCNA analysis. Potential biomarker genes in ACC were identified through association between gene sets and OS. In the current study, gene expres- sion matrices for the TCGA-ACC cohort and the GEO-GSE76021 cohort were used as input data. The “WGCNA” package in R

software was used for WGCNA processing. An adjacency ma- trix and a topological overlap matrix were constructed. After that, we constructed a gene dendrogram, and performed mod- ule identification with a dynamic tree cut. The correlation be- tween OS and genomic data was then calculated. The most predominant module associated with OS was selected as the key module, and genes in the key module were used in the following analysis.

Survival Analysis

Univariate Cox analysis was used to screen survival-related genes. Hazard ratio (HR) was used to classify genes as either protective or deleterious. These genes were selected as po- tential prognostic genes. A multivariate Cox regression mod- el including gender, age, tumor stage, NDRG Family Member 4 (NDRG4), and CDC28 Protein Kinase Regulatory Subunit 2 (CKS2) was used to identify independent prognostic factors for OS in ACC. Kaplan-Meier survival analysis and log-rank test were used to validate the prognostic value of target genes.

Differential Expression Analysis

For differential expression analysis, long-term survivors were defined as those with OS >5 years, and short-term survivors were defined as those with OS <3 years. Differentially expressed genes (DEGs) were determined by the “Limma” package mod- erated t test, and genes with P value <0.001 and fold change >2 were regarded as DEGs [26].

Establishment of regression Model and Construction Of Risk Score

The risk score for predicting postoperative survival was calcu- lated based on a linear combination of tumor stage and gene expression in the TCGA-ACC cohort by logistic regression. The following calculation formula was used for the analysis:

Risk score= Σ Ni=1 (Expi*Coei)

N, Coei, and Expi represented gene number, coefficient value, and level of gene expression, respectively. Receiver operating characteristic (ROC) analysis was used to evaluate the accu- racy of the prognostic model.

Gene Set Enrichment Analysis

Gene set enrichment analysis (GSEA) was performed us- ing the molecular signatures database (MSigDB) (http:// www.broadinstitute.org/gsea/msigdb). The “clusterProfiler” R package was used during the process. Significantly differ- ent pathways were defined as P< 0.05 in the gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG)

databases. Enrichment scores and P values were based on 1000 permutations.

Tumor Microenvironment Analysis

We applied xCell (http://xCell.ucsf.edu/), a gene signature-based deconvolution method, to infer 64 immune and stromal cell types for ACC microenvironment analysis [27].

Statistical Analysis

Wilcoxon signed ranks test or t test was used for analysis of continuous variables, as appropriate. Chi-square test, Fisher’s exact test, or Cochran-Mantel-Haenszel test was applied for categorical variables. Correlational analyses were performed by Spearman’s correlation test. Two-tailed P value <0.05 was considered as statistically significant for all statistical analyses.

Results

Identification of Potential Prognostic Genes in the TCGA-ACC Cohort

Figure 1 shows the study design. We first performed weight- ed correlation network analysis to identify prognostic genes in the TCGA-ACC cohort. OS was set as the phenotype. Expression matrices for 79 samples were used for WGCNA construction. Figure 2A shows the cluster dendrogram of the samples in the TCGA-ACC cohort. After quality assessment for each ex- pression matrix, R2=0.9 was selected to ensure a scale-free network (Figure 2B). We also set a threshold of 0.4 to merge similar modules. After that, a total of 15 modules were iden- tified (Figure 2C). Genes in the grey module were removed in the subsequent analysis. Next, we calculated eigengenes for all of the modules and clustered them based on their correla- tion (Figure 2D, 2E). A module eigengene heatmap showed that the 7 modules were mainly divided into 3 clusters (Figure 2D). A network heatmap was also performed to analyze the inter- actions between the modules (Figure 2E). The results showed that the skyblue2 module was relatively independent from the other modules, with a high-scale independence of transcrip- tomic expression. Figure 2F shows a heatmap of the correla- tion between module eigengenes and patients’ survival in the TCGA-ACC cohort. Close relationships are represented by red color. We found that the skyblue2 module was significantly correlated with OS (Figure 2F, P=9×10-8). Taken together, the skyblue2 module was selected as the main prognostic module.

We further performed differential expression analysis to de- fine 3769 DEGs between long-term survivors (OS >5 years) and short-term survivors (OS <3 years) in the TCGA-ACC co- hort (Figure 2G). Moreover, we performed a univariate Cox

Figure 2. Identification of prognostic genes in the TCGA-ACC cohort by WGCNA. (A) Cluster dendrogram of samples in the TCGA-ACC cohort to detect outliers. The dendrogram branches represent the clustered samples. (B) Soft-threshold power analysis of the scale-free fit index (upper) and the mean connectivity (lower). (C) Dendrogram of gene clustering and module assignment. Modules are represented by different colors. (D) Heatmap of adjacencies in the eigengene network. Higher adjacency is represented by red color. (E) Interaction of co-expressing modules. Higher connectivity is represented by light color. (F) Heatmap of the correlation between module eigengenes and patients' survival. Close relationships are represented by red color. (G) Volcano plot of DEGs between long-term survivors and short-term survivors in the TCGA-ACC cohort. (H) Venn diagram of the intersection of DEGs between long-term survivors and short-term survivors, prognostic genes according to univariate Cox regression model, and the WGCNA-skyblue2 module. WGCNA - weighted correlation network analysis; DEGs - differentially expressed genes.

A

B

D

Sample denrogram and trait heatmap

Scale independence

121314151617181820

Scale free topology model fit, signed R2

1.0

1.0

160

0.8

0.8

0.6

3

0.4

0.6

Height

140

MEp lum

MEbrovin

MEcoral2

120

TOGA-OR-A5J8-01A

TOGA-OR-ASKT-01A TOGA-OR-ASKED

ICGA-OR-ASJG-01AOHASLO-01A

TOGA-OR- ASLO-OTA

KOGA- OR-A 515-01A

ICGA-OR-ASL3-01A

0.2

0.4

2

MEantiquevihite4

MEmediumorchid

MElightsteelblue1

MEdarkred

MEskyblue1

MEI avenderblush

MEbrown4

MEbarkseagreen

MEcoral

MEpurple.

MEskyblue?

0.0

1

100

KOGA-OR-SJ8-01A KGA-P6-ASOG-01A

TOGA-OR-ASLC-01A TOG A-OR-ASK7-01A

TOGA-OR-ASK9-01A

TOGA-OR-ASLE-01A

TOGA-08-15JM-01A

5

10

15

20

80

TOGA-OR-ASLK- 01A

TEGA -OR- ASAN-01A

TOG A-OR-A511-01.

ICIGA -OR- ASJV-01A

KOGA-OH-A SJ0-01A

1.0

TOGI-OH-ASJR-01A

HUN OR ASLH- 01A

KOGA-OR-ASAL-01A

TOCA-PR- ASHB-01A

LOGAN -OR -ASKS-01A TOGA-OR-ASLMOTA

ICGA-OR-A SKY-01

GAOR ASTE OL

IEGA -OR- ASJF-01

TOGA-OR ASI&LOLA

KOGA-OR-ASKW-01

TOGA-OR-ASI7-01A

TOGA-OR-ASK2-01A

TOGA-OR -ASJE-01A

Soft threshold (power) Scale independence

3000

1

0.8

Mean connectivity

0.6

2000

0.4

OS time

1000

2

0.2

500

3

0

0.0

5 6 7 8 9 1011121314151617181820

5

10

15

20

C

Cluster dendogram

Soft threshold (power)

E

Network heatmap plot, selected genes

1.0-

0.8

Height

0.6-

0.4

0.2

Dynamic tree cut

Merged dynamic

TCGA-COX

TCGA-DEGS

F

G

H

2787

2949

0

MEplum

0.17

12.5

MEbrown

(0.3)

0.16

Down

.

820

MEcoral2

(03)

1.0

Stable

Up

1837

0

MEantiquewhite4

-0.066

10.0

MEmediumorchid

(0.7)

MElightsteelblue 1

0.19 (0.2)

MEdarkred

0.18 (0.3)

0.5

2751

-log10 (p-value)

7.5

MEskyblue1

0.15

MEbrown4

(0.3)

0.0

WGCNA-skyblue2

MEdarkseagreen4

5.0

Size of each list

0.44

MElavenderblush3

10.004)

8393

(0.2)

4196.5

8393

MEcoral1

0.47

-0.5

(0.002)

2.5

0

3769

5408

MEpurple

0.63

MEskyblue2

(9e-06]

TCGA-COX TCGA-DEXs WGCNA-skyblue2 Number of lements: specific (1) or shared by 2,3, … lists

0.72

MEgrey

(ge-08)

-0.048

(0.8)

-1.0

0.0

4786

5538

OS time

-2.5

0.0

2.5

3

2

1

log2 (fold change)

DATABASE ANALYSIS

Figure 3. Identification of prognostic genes in the GSE76021 cohort. (A) Cluster dendrogram of samples in the GSE76021 cohort. The dendrogram branches represent the clustered samples. (B) Soft-threshold power analysis of the scale-free fit index (upper) and the mean connectivity (lower). (C) Dendrogram of gene clustering and module assignment. Modules are represented by different colors. (D) Heatmap of adjacencies in the eigengene network. Higher adjacency is represented by red color. (E) Interaction of co-expressing modules. Higher connectivity is represented by light color. (F) Heatmap of the correlation between module eigengenes and patients' survival. Close relationships are represented by red color. (G) Volcano plot of DEGs between long-term survivors and short-term survivors in the GSE76021 cohort. (H) Venn diagram of the intersection between DEGs, prognostic genes according to Cox analysis, and the WGCNA-plum3 module. WGCNA - weighted correlation network analysis; DEGs - differentially expressed genes.

A

B

D

Sample denrogram and trait heatmap

Scale independence

1.1

Scale free topology model fit, signed R2

0.8

7 8 91011121914151617181820

80

0.9

75

GSM 1972921

0.6

Height

GSM1972924

0.7

70

GSM1972939

GSM197233

0.4

0.5

MEnavajowhite1

65

GSM11972927

GSM11972931

GSM1972937

GSM19/2935

0.2

MEbrovinz

MEbrown

MEgreen

MEdarlorange

MEdarkturqu ob

MEdarkgry

MBightyanl

ROPA

MEblack

MEplum3

MElightsteelbluet

ME avenderblush2

MEorangered4

MEantiquewhite4

MEbluež

1

60

GSM1972926

GSM1972928

GSM 1972944

GSM1972919

GSM1972929

GSM1972938

GSM1972942

5

10

15

55

Soft threshold (power)

20

50

GSM1972920

GSM1972922

GSM1972925

GSM1972923

GSM1972940

GSM1972934

GSM1972930

GSM1972947

3000

Scale independence

1.0

45

1

0.8

Mean connectivity

2000

0.6

0.4

OS time

1000

0.2

500-

0-

0.0

5 6 7 8 9 1011121314151617181820

5

Soft threshold (power)

10

15

20

C

Cluster dendogram

E

Network heatmap plot, selected genes

1.00-

0.95-

0.90

Height

0.85

0.80

0.75

0.70-

0.65-

Dynamic tree cut

Merged dynamic

GEO-COX

GEO-DEGS

F

G

H

774

255

144

MEplum

0.21

(0.3)

Down

MEbrown

5

-0.29

MEcoral2

(0-2)

1.0

Stable

-0.17

Up

25

6

MEantiquewhite4

(944)

0_27

00.2)

MEmediumorchid

0-067

4

MElightsteelblue1

02

302

00.4)

0.5

MEdarkred

0.44

[0.02)

MEskyblue1

0.36

-log 10 (p-value)

[0.03)

MEbrown4

0.13 00.6)

WGCNA-plum3

0,34

0.0

MEdarkseagreen4

(0.1)

2

Size of each list

MElavenderblush3

-0.29 (0.2)

1059

-0.26

529.5

1095

MEcoral1

00.2)

0.076

410

338

MEpurple

(0.7)

0.5

0

-0.28

TCGA-COX

MEskyblue2

TCGA-DEXs WGCNA-skyblue2

00.2)

-0.47

0

Number of lements: specific (1) or shared by 2,3, … lists

MEgrey

[0.02)

[Be-040)

286

1220

OS time

1.0

-2

-1

0

1

2

3 (5) 2

1

log2 (fold change)

Figure 4. Identification of NDRG4/CKS2 as key prognostic genes in ACC. (A) Venn diagram of the intersection of prognostic genes in the TCGA-ACC cohort and the GSE76021 cohort. (B) Multivariate Cox regression model for overall survival in ACC including gender, age, tumor stage, NDRG4 expression, and CKS2 expression. (C) NDRG4 gene expression in ACC according to different stages. (D) CKS2 gene expression in ACC according to different stages. ACC - adrenocortical carcinoma.

A

WGCNA-GEO

B

WGCNA-TCGA

Hazard ratio

31

199

14

TCGA-differ

1.32

11

373

Gender

(n=79)

(0.56-3.1)

H

4

0.53

2435

169

11

13

Age

(n=79)

1.01

(0.98-1.0)

0.549

19

1666

65

2

1.80

72

8

Stage

(n=79)

78

(1.14-2.8)

0.012*

11

14

GEO-differ

393

NDRG4

(n=79)

0.61 (0.46-0.8)

<0.001 ***

65

5090

485

GEO-COX

CKS2

(n=79)

2.52

(1.38-4.6)

0.003 **

Events: 27; Global p-value (Log-Rank): 3.0303e-11 aIC: 159.51; Concordance index: 0.88

TCGA-COX

0.5

1

2

5

C

NDRG4

CKS2

Size of each list

0.94

0.0092

0.0087

0.00011

8393

8393

0.23

12.5

0.71

5408

0.0043

1.9e-05

4196.5

Gene expression

10

0.093

10.0

0.24

1969

0.31

0.038

338

454

1059

0

:

WGCNA-TCGA

7.5

WGCNA-gGEO

TCGA-differ

GEO-differ

TCGA-COX

GEO-COX

5

:

Number of lements: specific (1) or shared by 2,3, … lists

5.0

2654

8568

2.5

6 (2)

5 (132)

2

1

0

T1

T2

T3

T4

T1

T2

T3

T4

4 (154)

3 (819)

regression model to define 8393 potential prognostic genes (Figure 2H). Finally, 820 intersection genes were defined be- tween the DEGs, Cox regression prognostic genes, and the WGCNA-skyblue2 module genes.

Identification of Potential Prognostic Genes in the GSE76021 Cohort

We downloaded clinicopathologic and transcriptomic data from the GEO dataset GSE76021, which consisted of data from 29 ACC patients. We also set OS as the phenotype for WGCNA. A cluster dendrogram of the samples in the GSE76021 cohort is illustrated in Figure 3A. After quality assessment for the ex- pression matrix, R2=0.9 was selected to ensure a scale-free net- work (Figure 3B). We also set a threshold of 0.4 to merge sim- ilar modules. After that, a total of 16 modules were identified (Figure 3C). Next, we calculated eigengenes for all modules and clustered them based on their correlation (Figure 3D, 3E). The modules were mainly divided into 2 clusters (Figure 3D), and the interactions among all of the modules were detect- ed by network heatmap (Figure 3E). In the end, the plum3 module was defined as relatively independent from the other

modules, with a high-scale independence of transcriptomic ex- pression. Moreover, a close relationship was found between the plum3 module and the patients’ OS in the GSE76021 co- hort (Figure 3F). Taken together, the plum3 module was se- lected as the main prognostic module in the GSE76021 cohort.

Differential expression analysis was further performed in the GSE76021 cohort, and 410 DEGs were found between long- term survivors (OS >5 years) and short-term survivors (OS <3 years) (Figure 3G). Moreover, univariate Cox analysis was per- formed, and 1059 potential prognostic genes were identified (Figure 3H). Finally, 5 intersection genes were defined between the DEGs, Cox regression prognostic genes, and WGCNA-plum3 module genes in the GSE76021 cohort.

Identification of NDRG4 and CKS2 as Prognostic Genes for ACC

Figure 4A shows a Venn diagram of the intersection of prog- nostic genes in the TCGA-ACC cohort and the GSE76021 co- hort. We next constructed a multivariate Cox regression mod- el for OS in ACC, including gender, age, tumor stage, NDRG4,

DATABASE ANALYSIS

Figure 5. Association between NDRG4/CKS2 expression and survival of ACC patients. (A) Kaplan-Meier survival curve of overall survival in the TCGA-ACC cohort according to NDRG4 and CKS2 expression. (B) Survival curve of overall survival in the GSE76021 cohort according to NDRG4 and CKS2 expression. (C) Survival curve of disease-free survival in the TCGA-ACC cohort according to NDRG4 and CKS2 expression. (D) Survival analysis according to NDRG4 and CKS2 expression in ACC patients treated by mitotane therapy. (E, F) Receiver operating characteristic analysis of the tumor stage, CKS2 expression, NDRG4 expression, and the combined risk score for predicting 3-year survival (E) and 5-year survival (F) in the TCGA-ACC cohort. ACC - adrenocortical carcinoma.

A

TCGA-ACC cohort

B

GSE76021 dataset

NDRG4 level + High+Low

CKS2 level

+ High+Low

NDRG4 level + High+Low

CKS2 level

+High+Low

1.00

1.00

1.00-

1.00

Survival probability

0.75

Survival probability

0.75

Survival probability

0.75

Survival probability

0.75

0.50

0.50

0.50

0.50

0.25

p<0.001

0.25

p<0.001

0.25

p=0.002

0.25

p

0.001

0.00

0.00

0.00

0.00

0

T

2

3

4

5

6

7

8

9101112

0

1

2

3

4

5

6

7

8

9

10 11 12

0

2.

3

1

6

8

9 1011 12 131415 16

0

2

4

5 6

8

9 9 10

01112131415 16

Time (years)

Time (years)

Time (years)

Time (years)

C

Disease-free survival

D

Mitotane therapy

NDRG4 +High+Low

CKS2 +High+Low

NDRG4

High+Low

CKS2 +High+Low

1.00

1.00

1

1.00-

1.00-

Survival probability

0.75

Survival probability

0.75

Survival probability

0.75

Survival probability

0.75

0.50

0.50

0.50

0.50

0.25

p<0.001

0.25

p<0.001

0.25

p<0.001

0.25

p<0.001

0.00

1

2

7

8

9101112

0.00

0.00

0.00

0

3

4

5

6

0

1

2

3

4

5

6 7

8

9 1011 12

0

1

2

3

4

5

6

7

8

9101112

0

1

2

3

4 5

6

7

8

9 101112

Time (years)

Time (years)

Time (years)

Time (years)

E

3-year survival

F

5-year survival

100

100

80

80

Sensivity

60

Sensivity

60

40

40

20

AUC

P value

Stage

0.804

reference

20

AUC

P value

0.891

CKS2

0.882

0.2091

Stage

reference

0.1182

CKS2

0.892

0.9763

NDRG4

0.906

NDRG4

0.866

0.6718

0

Risk score

0.974

0.0016

0

Risk score

0.969

0.0313

0

20

40

60

80

100

0

20

40

60

80

100

100-specificity

100-specificity

Figure 6. Association between NDRG4/CKS2 expression and homologous recombination/mismatch repair pathways. (A) GSEA analysis of enriched pathways in NDRG4 low-expression samples. (B) Heatmap of genes in the homologous recombination pathway and NDRG4 expression. (C) Association between NDRG4 and BRCA1/BRCA2/EXO1/MSH2 expression in the TCGA-ACC cohort. (D) GSEA analysis of enriched pathways in CKS2 high-expression samples. (E) Heatmap of genes in mismatch repair pathways and CKS2 expression. (F) Association between CKS2 and BRCA1/BRCA2/EXO1/MSH2 expression in the TCGA-ACC cohort. GSEA, gene set enrichment analysis.

A

NDRG4

B

NDRG4 low

NDRG4 high

cell_list

- KEGG_CELL_CYCLE

CHEK2

4

BADSNC

0.6

- KEGG_DNA_REPLICATION

BARD1

2

CHECKT

0

Enrichment score

- KEGG_HEDGEHOG_SIGNALING_PATHWAY

- KEGG_HOMOLOGOUS_RECOMBINATION

-2

0.4

-4

.

FANCDZ

FANCA

0.2

ELM

FANCE

FANCF

FANCL

BAD52 ATM

0.0

MRE11A

ATR

FANCM

BAD518 COK12

PALB2

BADSO

high expression

low expression

C

0.29, p = 0.0097

R =- 0.41, p= 2e-04

R = - 0.56, p= 1.2e-07

R= · 0.44, p = 5.4e-05

BRCA1

BRCA2

EXO1

MSH2

NDRG4

NDRG4

NDRG4

NDRG4

D

E

CKS2

CKS2 low

CKS2 high

0.8

cel_list

4

KEGG_DNA_REPLICATION

MLH3

PMSZ

2

Enrichment score

- KEGG_MISMATCH_REPAIR

MSH3

0.6

- KEGG_P53_SIGNALING_PATHWAY

RFCT

POLD4

0

- KEGG_SPLICEOSOME

MLH1

RPA1

-2

0.4

SSEPT

RPA3

4

RPA4

POMA

RFC4

0.2

POLD3

REC2

RPAZ

MSH2

0.0

MISHG

RFC3

RFCS

LIGT

POLD1

YCOLA CREASKK01A

TOGA OR ASI

TOO.A. OR ASKEDIN

QUIASPIORA TOGA OFLASJBL.DIA

TOGA OR ASJYCIA

CIEASJFORA

TOO.A. DR ALICIA

TOGA GRASNO DIA

TODA CILASJ7.01A

TOGA OR ASKEDIA

TOUA OR ASKCZOŁA

TOG.A. ORLASKYDIA

high expression

low expression

F

R=0.46, p = 2.3e-05

R= 0.55, p = 1.9e-07

R=0.68, p < 2.2e-16

R= 0.46, p = 2.48-05

BRCA1

BRCA2

EXO1

MSH2

CKS2

CKS2

CKS2

CKS2

DATABASE ANALYSIS

4

A

p=0.002

0.5

p<0.001

0.4

p<0.001

p=0.008

Fraction

0.3

p<0.001

p=0.003

0.2

P<0.001

p<0.001

p<0.001

0.1

p=0.089

p=0.006

p=0.036

p=0.001

p<0.001

p=0.633

p=0.284

p=0.577

0.0

CD4+ T-cells

CD8+ T-cells

CD8+ Tcm

CD8+ Tem

Fibroplast

MSC

Macrophages

Macrophages M1

Macrophages M2

NK cells

NKT

Plateles

Th1 cells

Th2 cells

Tregs

CDC

DC

B

R=0.47,p=1e-05

R=0.29,p=0.0092

0.20

R=0.29,p=0.0083

R=0.59,p=8.4e-09

0.15

0.10

0.06

CD8+T cells

0.15

0.10

DC

0.05

NKT

0.10

Tregs

0.04

0.05

0.05

0.02

0.00

0.00

0.00

0.00

0

2

4

6

8

0

2

4

6

8

0

2

4

6

8

0

2

4

6

8

NDRG4

NDRG4

NDRG4

NDRG4

0.15

R=0.56, p=9.2e-08

0.3

R=0.39, p=0.00044

0.075

‘R=0.23,p=0.042

0.3

R=0.42,p=0.00011

0.10

Th1 cells

Th2 cells

0.2

Macrophages

0.050

Fibroblasts

0.2

0.05

0.1

0.025

0.1

0.00

0.0

0.000

0.0

0

2

4

6

8

0

2

4

6

8

0

2

4

6

8

0

2

4

6

8

NDRG4

NDRG4

NDRG4

NDRG4

C

P<0.001

0.5

p<0.001

0.4

p=0.002

P<0.001

Fraction

0.3

0.2

p<0.001

P<0.001

p=0.002

P<0.001

p<0.001

0.1

p=0.120

p=0.016

p=0.228

p<0.001

p<0.001

p=0.301

p=0.308

p=0.002

0.0

CD4+T-cells

CD8+T-cells

CD8+ Tcm

CD8+ Tem

Fibroplast

MSC

Macrophages

Macrophages M1

Macrophages M2

NK cells

NKT

Plateles

Th1 cells

Th2 cells

Tregs

DC

CDC

4

6

8

6 8

4

6

8

4

6

8

Indexed in: [Current Contents/Clinical Medicine] [SCI Expanded] [ISI Alerting System] [ISI Journals Master List] [Index Medicus/MEDLINE] [EMBASE/Excerpta Medica] [Chemical Abstracts/CAS]

Figure 7. Association between NDRG4/CKS2 expression and immunological status in ACC. (A) Fraction of immune cell infiltration in the NDRG4 high-expression and low-expression samples according to xCell cell type enrichment analysis. (B) Correlation between immune cell infiltration and NDRG4 expression. (C) Fraction of immune cell infiltration in the CKS2 high-expression and low-expression samples according to xCell analysis. (D) Correlation between immune cell infiltration and CKS2 expression. ACC - adrenocortical carcinoma.

D

R=0.51,p=2e-06

0.075

0.15

R =- 0.59, p=7.7e-09

0.20

R =- 0.37,p=0.00077

R =- 0.65, p=9.3e-11

0.10

0.15

CD8+T cells

0.050

0.10

DC

0.05

NKT

0.10

Tregs

0.025

0.05

0.05

0.00

0.00

0.000

0.00

4

6

8

4

6

8

4

6

8

4

6

8

CKS2

CKS2

CKS2

CKS2

0.15

R=0.61,p=2.9e-09

0.3

R=0.55,p=1.6e-07

0.075

R=0.38,p=0.00056

0.3

R =- 0.41,p=0.00014

0.10

0.2

Macrophages

0.05

0.050

0.2

Th 1 cells

Th2 cells

Fibroblasts

0.1

0.00

0.0

0.025

0.1

0.0

-0.1

0.000

4

6

8

4

6

8

4

6

8

4

6

8

CKS2

CKS2

CKS2

CKS2

and CKS2 (Figure 4B). In this multivariate Cox regression model, NDRG4 and CKS2 expression were both independent- ly associated with the patients’ OS (NDRG4: HR 0.61, 95% CI 0.46-0.80, P<0.001; CKS2: HR 2.52, 95% CI 1.38-4.60, P=0.003; Figure 4B). Moreover, NDRG4 expression decreased when tu- mor stage increased (Figure 4C). In contrast, CKS2 expression increased when tumor stage increased. The results indicat- ed that NDRG4 and CKS2 expression are potential prognos- tic genes in ACC, and are associated with ACC progression.

We also performed Kaplan-Meier survival analysis for the ACC patients. Kaplan-Meier survival curves and log-rank analysis showed that high expression of NDRG4 was associated with better survival in both the TCGA-ACC cohort (P<0.001; Figure 5A) and the GSE76021 cohort (P=0.002; Figure 5B). On the other hand, CKS2 high expression was associated with worse sur- vival in both the TCGA-ACC cohort (P<0.001; Figure 5A) and the GSE76021 cohort (P<0.001; Figure 5B). For disease-free survival, Kaplan-Meier survival analysis also identified NDRG4 (P<0.001) and CKS2 (P<0.001) expression as prognostic factors (Figure 5C). Moreover, NDRG4 (P<0.001) and CKS2 (P<0.001) expression could also be prognostic factors for patients treat- ed with mitotane therapy (Figure 5D).

We further performed ROC analysis to identify whether CKS2 and NDRG4 could be used as predictors for 3-year survival and 5-year survival of ACC patients. We constructed a combined risk score by logistic regression modelling for ACC survival

according to tumor stage, NDRG4 expression, and CKS2 ex- pression. We found that this combined risk score could be used for prediction of 3-year survival in ACC, and this worked better than tumor stage alone (AUC=0.974 vs AUC=0.804, P=0.0016, Figure 5E). The combined risk score could also be used for pre- diction of 5-year survival, a stronger prediction than for tumor stage alone (AUC=0.969 vs AUC=0.891, P=0.0313; Figure 5F).

In order to clarify biologic pathways related to NDRG4 and CKS2 expression, we further performed GSEA analysis in the TCGA- ACC cohort. We found several pathways enriched in NDRG4 low-expression samples including the cell cycle pathway, the DNA replication pathway, the hedgehog signaling pathway, and the homologous recombination pathway (Figure 6A). Moreover, expression of genes related to the homologous recombina- tion pathway was mostly elevated in the NDRG4 low-expres- sion samples (Figure 6B). In addition, a negative correlation was found between NDRG4 expression and homologous re- combination genes including BRCA1, BRCA2, EXO1, and MSH2 (Figure 6C).

We also performed GSEA analysis between low and high ex- pression levels of CKS2. We found DNA replication mismatch repair, p53 signaling pathway, and spliceosome pathway genes were enriched in CKS2 high-expression samples (Figure 6D). Moreover, expression of genes related to DNA mismatch repair

pathways was mostly elevated in the CKS2 high-expression samples (Figure 6E). In addition, a positive correlation was found between NDRG4 expression and mismatch repair genes such as BRCA1, BRCA2, EXO1, and MSH2 (Figure 6F).

Immune Cell Infiltration Associated with NDRG4 and CKS2 Expression

Immunotherapy has been recently applied in multiple cancer types. We aimed to discover whether NDRG4 and CKS2 ex- pression was associated with immune cell infiltration in ACC. By using xCell cell type enrichment analysis, we identified im- mune cell infiltration in the TCGA-ACC samples. The infiltra- tion of CD8+ T cells, dendritic cells, natural killer T cells, regu- latory T cells, macrophages, and fibroblasts was significantly more abundant in NDRG4 high-expression samples, while the infiltration of Th1 and Th2 cells was more abundant in NDRG4 low-expression samples (Figure 7A). Correlation analysis also identified a positive correlation between NDRG4 expression and infiltration of CD8+ T cells, dendritic cells, natural killer T cells, regulatory T cells, macrophages, and fibroblasts; and a negative correlation between NDRG4 expression and infiltra- tion of Th1 and Th2 cells (Figure 7B).

We also analyzed the relationship between CKS2 expression and immune cell infiltration. The infiltration of CD8+ T cells, dendritic cells, natural killer T cells, regulatory T cells, macro- phages, and fibroblasts was significantly more abundant in CKS2 low-expression samples, while the infiltration of Th1 and Th2 cells was more abundant in CKS2 high-expression sam- ples (Figure 7C). Correlation analysis also identified a nega- tive correlation between CKS2 expression and infiltration of CD8+ T cells, dendritic cells, natural killer T cells, regulatory T cells, macrophages, and fibroblasts; and a positive correlation between CKS2 expression and infiltration of Th1 and Th2 cells (Figure 7D). Our results suggested immune cell infiltration dif- fers according to NDRG4 and CKS2 expression.

Discussion

ACC is an aggressive cancer with heterogeneous outcomes, but its molecular mechanisms have not yet been clarified. Recently, genomic technologies have been applied for the discovery of mechanisms of ACC. In the present study, we applied transcrip- tional analysis of available transcriptomic data from 2 data- sets of ACC. We identified NDRG4 and CKS2 as key prognos- tic genes for ACC, with potential predictive value for mitotane therapy as well. We further identified mismatch repair path- ways as dominant biologic pathways associated with CKS2. Moreover, immune cell infiltration in ACC differed according to NDRG4 and CKS2 expression.

NDRG4 is a member of the N-myc downregulated gene family, which belongs to the alpha/beta hydrolase superfamily [28]. The protein encoded by this gene is a cytoplasmic protein that is required for cell cycle progression and survival [29], and may be involved in the regulation of mitogenic signaling in vascu- lar smooth muscles cells [30]. Research into the connection between NDRG4 and cancer is gaining more and more atten- tion, although conflicting results have been reported. NDRG4 can show either tumor-suppressive or oncogenic function, de- pending on the tumor types [31-34]. Currently, the function of NDRG4 in ACC has not been determined. Our study found a relationship between high expression of NDRG4 and favor- able survival in ACC (Figure 5A-5C), suggesting that it may be a potential tumor suppressor. Since NDRG4 is required for tu- mor cell cycle regulation [29], and cell cycle regulation plays a crucial role in ACC progression [12], we believe this could be a potential mechanism underlying the correlation between NDRG4 and ACC development. Moreover, Agosta et al (2018) found that NDRG4 may be a potential target of miR-139-5p, and that the miR-139-5p/NDRG4 pathway promotes ACC ag- gressiveness by mediating epithelial-to-mesenchymal transi- tion, which results in tumor cell invasiveness and anchorage- independent growth [35]. However, this finding needs further in vitro and in vivo validation.

CKS2 is a protein-coding gene that binds to the catalytic sub- unit of the cyclin-dependent kinases and is essential for their biological function [36]. CKS2 is involved in the process of tu- morigenesis of gastric cancer [37], prostate cancer [38], blad- der cancer [39] and others. Our study also found an associ- ation between high expression of CKS2 and poor survival in ACC (Figure 5A-5C). Moreover, we found CKS2 expression to be correlated with elevated expression of mismatch repair genes such as BRCA1, BRCA2, EXO1, and MSH2 (Figure 6D-6F). Since CKS2 overexpression has been found to be able to override the DNA damage response barrier triggered by activated oncopro- teins [40], we believe this could be a potential mechanism un- derlying the correlation between CKS2 and survival of ACC pa- tients. Moreover, through in vitro experiments, Chehade et al found that miR-7, which participates in cell cycle arrest in ACC, acts as a key regulator of CKS2 [41]. According to these stud- ies, we speculate that miRNAs might play an important role in the regulation of NDRG4 and CKS2, resulting in tumor cell proliferation and aggressiveness, as well as ACC progression. However, further in vitro and in vivo functional studies are still needed. In vitro studies needed include gene overexpression and CRISPR knock-out in tumor cells, and evaluation of tumor cell growth, invasion, and other related biological processes. In vivo studies could be performed on tumor-bearing mice, includ- ing evaluation of tumor growth and mouse survival. Moreover, the functional status of immune cells in tumor-bearing mice should also be evaluated, since our study also suggested a re- lationship between CKS2/NDRG4 and intratumoral immunity.

Mitotane is recommended as systemic therapy for metastat- ic or unresectable ACC, and also as adjuvant therapy in high- risk postoperative patients [20], but not all patients respond to this drug. The development of robust predictive biomarkers for mitotane therapy efficacy may have potential clinical sig- nificance. Our study identified NDRG4 and CKS2 as predictive biomarkers for mitotane therapy success in ACC (Figure 5D), which may be helpful in clinical management. However, further validation studies in larger, prospective cohorts are still needed.

Sequence-based analyses, including genomic, transcriptomic, and epigenomic profiling, have been applied recently in ACC studies [18]. Such studies have identified novel driver genes and molecular classifications for ACC, which may guide fur- ther functional studies [18]. Our study integrated multiple da- tasets from previous studies, which provided a more compre- hensive view of the pathogenesis of ACC.

The present study still has some limitations. The limited sam- ple size is the major limitation. Further validation studies in larger, prospective cohorts are needed. Moreover, the meth- odology of the current study could not clarify the underlying molecular mechanism of the 2 prognostic genes. Future func- tional and mechanical studies are needed to fully understand the underlying mechanisms. In addition, the study indicated

References:

1. Crona J, Beuschlein F. Adrenocortical carcinoma - towards genomics guid- ed clinical care. Nat Rev Endocrinol, 2019;15(9):548-60

2. Datta J, Roses RE. Surgical management of adrenocortical carcinoma: An evidence-based approach. Surg Oncol Clin N Am, 2016;25(1):153-70

3. Else T, Kim AC, Sabolch A, et al. Adrenocortical carcinoma. Endocr Rev, 2014;35(2):282-326

4. Erickson LA, Rivera M, Zhang J. Adrenocortical carcinoma: Review and up- date. Adv Anat Pathol, 2014;21(3):151-59

5. Fassnacht M, Johanssen S, Quinkler M, et al. Limited prognostic value of the 2004 International Union Against Cancer staging classification for ad- renocortical carcinoma: Proposal for a Revised TNM Classification. Cancer, 2009;115(2):243-50

6. Margonis GA, Kim Y, Prescott JD, et al. Adrenocortical carcinoma: Impact of surgical margin status on long-term outcomes. Ann Surg Oncol, 2016;23(1):134-41

7. Beuschlein F, Weigel J, Saeger W, et al. Major prognostic role of Ki67 in lo- calized adrenocortical carcinoma after complete resection. J Clin Endocrinol Metab, 2015;100(3):841-49

8. Xiao H, Xu D, Chen P, et al. Identification of five genes as a potential bio- marker for predicting progress and prognosis in adrenocortical carcinoma. J Cancer, 2018;9(23):4484-95

9. Lippert J, Appenzeller S, Liang R, et al. Targeted molecular analysis in adre- nocortical carcinomas: A strategy toward improved personalized prognos- tication. J Clin Endocrinol Metab, 2018;103(12):4511-23

10. Sbiera S, Sbiera I, Ruggiero C, et al. Assessment of VAV2 expression refines prognostic prediction in adrenocortical carcinoma. J Clin Endocrinol Metab, 2017;102(9):3491-98

11. Subramanian C, Cohen MS. Over expression of DNA damage and cell cycle dependent proteins are associated with poor survival in patients with ad- renocortical carcinoma. Surgery, 2019;165(1):202-10

12. Pereira SS, Monteiro MP, Bourdeau I, et al. Mechanisms of endocrinol- ogy: Cell cycle regulation in adrenocortical carcinoma. Eur J Endocrinol, 2018;179(2):R95-110

a relationship between immunogenic status and prognosis in ACC. Further studies are needed to clarify whether or not im- mune cell infiltration is involved in ACC pathogenesis and de- velopment, and whether immune cell infiltration could be a potential therapeutic target in ACC. The prognostic genes iden- tified in this preliminary study require validation by further in- terventional and functional studies.

Conclusions

This study applied a comprehensive transcriptomic approach to analyze available transcriptomic data to identify key prognostic genes in ACC. We identified NDRG4 and CKS2 as key prognos- tic genes in ACC that also may have potential predictive val- ue for mitotane therapy. We further identified DNA mismatch repair pathways as dominant biologic pathways associated with CKS2. Moreover, immune cell infiltration in ACC differed according to NDRG4 and CKS2 expression. These findings re- quire validation by further interventional and functional studies.

Conflicts of Interest

None.

13. Gara SK, Lack J, Zhang L, et al. Metastatic adrenocortical carcinoma dis- plays higher mutation rate and tumor heterogeneity than primary tumors. Nat Commun, 2018;9(1):4172

14. Doghman M, Cazareth J, Lalli E. The T cell factor/beta-catenin antagonist PKF115-584 inhibits proliferation of adrenocortical carcinoma cells. J Clin Endocrinol Metab, 2008;93(8):3222-25

15. Bernini GP, Moretti A, Viacava P, et al. Apoptosis control and proliferation marker in human normal and neoplastic adrenocortical tissues. Br J Cancer, 2002;86(10):1561-65

16. Fay AP, Signoretti S, Callea M, et al. Programmed death ligand-1 expression in adrenocortical carcinoma: An exploratory biomarker study. J Immunother Cancer, 2015;3:3

17. Caramuta S, Lee L, Ozata DM, et al. Clinical and functional impact of TARBP2 over-expression in adrenocortical carcinoma. Endocr Relat Cancer, 2013;20(4):551-64

18. Zheng S, Cherniack AD, Dewal N, et al. Comprehensive pan-genomic char- acterization of adrenocortical carcinoma [published correction appears in Cancer Cell, 2016;30(2):363]. Cancer Cell, 2016;29(5):723-36

19. Assié G, Letouzé E, Fassnacht M, et al. Integrated genomic characterization of adrenocortical carcinoma. Nat Genet, 2014;46(6):607-12

20. Lubitz JA, Freeman L, Okun R. Mitotane use in inoperable adrenal cortical carcinoma. JAMA, 1973;223(10):1109-12

21. Terzolo M, Baudin AE, Ardito A, et al. Mitotane levels predict the outcome of patients with adrenocortical carcinoma treated adjuvantly following rad- ical resection. Eur J Endocrinol, 2013;169(3):263-70

22. Hermsen IG, Fassnacht M, Terzolo M, et al. Plasma concentrations of o,p’DDD, o,p’DDA, and o,p’DDE as predictors of tumor response to mitotane in adre- nocortical carcinoma: Results of a retrospective ENS@T multicenter study. J Clin Endocrinol Metab, 2011;96(6):1844-51

23. Sbiera S, Leich E, Liebisch G, et al. Mitotane inhibits Sterol-O-Acyl Transferase 1 triggering lipid-mediated endoplasmic reticulum stress and apoptosis in adrenocortical carcinoma cells. Endocrinology, 2015;156(11):3895-908

DATABASE ANALYSIS

24. Leek JT, Johnson WE, Parker HS, et al. The sva package for removing batch effects and other unwanted variation in high-throughput experiments. Bioinformatics, 2012;28(6):882-83

25. Langfelder P, Horvath S. WGCNA: An R package for weighted correlation network analysis. BMC Bioinformatics, 2008;9:559

26. Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res, 2015;43(7):e47

27. Aran D, Hu Z, Butte AJ. xCell: Digitally portraying the tissue cellular hetero- geneity landscape. Genome Biol, 2017;18(1):220

28. Zhou RH, Kokame K, Tsukamoto Y, et al. Characterization of the human NDRG gene family: A newly identified member, NDRG4, is specifically ex- pressed in brain and heart. Genomics, 2001;73(1):86-97

29. Schilling SH, Hjelmeland AB, Radiloff DR, et al. NDRG4 is required for cell cycle progression and survival in glioblastoma cells. J Biol Chem, 2009;284(37):25160-69

30. Yang X, An L, Li X. NDRG3 and NDRG4, two novel tumor-related genes. Biomed Pharmacother, 2013;67(7):681-84

31. Ding W, Zhang J, Yoon JG, et al. NDRG4 is downregulated in glioblastoma and inhibits cell proliferation. OMICS, 2012;16(5):263-67

32. Xiao W, Zhao H, Dong W, et al. Quantitative detection of methylated NDRG4 gene as a candidate biomarker for diagnosis of colorectal cancer. Oncol Lett, 2015;9(3):1383-87

33. Chu D, Zhang Z, Zhou Y, et al. NDRG4, a novel candidate tumor suppressor, is a predictor of overall survival of colorectal cancer patients. Oncotarget, 2015;6(10):7584-96

34. Kotipatruni RP, Ren X, Thotala D, Jaboin JJ. NDRG4 is a novel oncogenic pro- tein and p53 associated regulator of apoptosis in malignant meningioma cells. Oncotarget, 2015;6(19):17594-604

35. Agosta C, Laugier J, Guyon L, et al. MiR-483-5p and miR-139-5p promote ag- gressiveness by targeting N-myc downstream-regulated gene family mem- bers in adrenocortical cancer. Int J Cancer, 2018;143(4):944-57

36. Spruck CH, de Miguel MP, Smith AP, et al. Requirement of Cks2 for the first metaphase/anaphase transition of mammalian meiosis. Science, 2003;300(5619):647-50

37. Kang MA, Kim JT, Kim JH, et al. Upregulation of the cycline kinase subunit CKS2 increases cell proliferation rate in gastric cancer. J Cancer Res Clin Oncol, 2009;135(6):761-69

38. Lan Y, Zhang Y, Wang J, et al. Aberrant expression of Cks1 and Cks2 con- tributes to prostate tumorigenesis by promoting proliferation and inhibit- ing programmed cell death. Int J Cancer, 2008;123(3):543-51

39. Chen R, Feng C, Xu Y. Cyclin-dependent kinase-associated protein Cks2 is as- sociated with bladder cancer progression. J Int Med Res, 2011;39(2):533-40

40. Liberal V, Martinsson-Ahlzén HS, Liberal J, et al. Cyclin-dependent kinase subunit (Cks) 1 or Cks2 overexpression overrides the DNA damage re- sponse barrier triggered by activated oncoproteins. Proc Natl Acad Sci USA, 2012;109(8):2754-59

41. Chehade M, Bullock M, Glover A, et al. Key MicroRNA’s and their targetome in adrenocortical cancer. Cancers (Basel), 2020;12(8):2198