ANALYSIS
Check for updates
Multi-omics-based molecular classification of adrenocortical carcinoma predicts response to immunotherapy and targeted treatments
Xingwei Jin1+, Xianjin Wang1+, Zhiyuan Wang2+, Baoxing Huang1, Xuejian Zhou1, Boke Liu1, Yuan Shao1* and Guoliang Lu1*
+Xingwei Jin, Xianjin Wang and Zhiyuan Wang have contributed equally to this work.
*Correspondence: Yuan Shao shaoyuan15@hotmail.com Guoliang Lu lgl_1111@163.com
1Department of UrologyRuijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
2Department of Anesthesiology Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Abstract
Adrenal cortical carcinoma (ACC) is a rare and highly aggressive malignant tumor with dismal outcomes. Once metastasis occurs, the 5-year survival rate falls below 15%. Current treatment options offer LIMited benefit for advanced disease, Largely due to the absence of well-defined therapeutic targets, and there is an urgent need to develop new molecular classifications to achieve precise treatment strategies. In this study, we integrated multi-omics data including transcriptome, epigenetic, and genomic variation profiles and applied 10 clustering algorithms, identifying two robust molecular subtypes of ACC: Multi-Omics ACC Consensus Subtyping (MACCS)1 and MACCS2. Biologically, MACCS1 exhibits a proliferation-driven phenotype, whereas MACCS2 displays an immune activation state. Drug sensitivity analysis further revealed that MACCS2 tumors were more responsive to immune checkpoint inhibitors, while MACCS1 showed sensitivity to antiangiogenic tyrosine kinase inhibition. Using a random forest algorithm, we identified HOXC11 as a key prognostic factor within MACCS1, with high expression associated with tumor progression. Functional assays confirmed that silencing HOXC11 significantly reduced the proliferation of ACC cells. Survival analysis showed that the prognosis of patients with MACCS1 had markedly worse outcomes compared to those with MACCS2. Collectively, this study provides a theoretical basis for the molecular classification of ACC and personalized precision treatment, such as immunotherapy and targeted therapy, and highlight HOXC11 as a potential therapeutic target.
Keywords Adrenocortical carcinoma, Multi-omics integrated analysis, Molecular classification, Targeted therapy
1 Introduction
Adrenal cortical carcinoma (ACC) is a rare but highly lethal endocrine malignancy [1, 2]. It is characterized by aggressive growth and early metastasis, leading to poor outcomes [3]. Nearly 75% of patients develop distant metastases, and the 5-year survival rate remains below 15%, creating a substantial health and economic burden worldwide [4]. Because ACC often presents with nonspecific symptoms, most patients are diagnosed
Discover
@ The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
at advanced stages. Radical surgery with adjuvant mitotane remains the standard treat- ment, while tumor heterogeneity and poorly understood resistance mechanisms limit therapeutic efficacy [5-8]. Therefore, establishing refined molecular subtypes and iden- tifying novel therapeutic targets are essential steps toward improving the clinical man- agement and extending the survival of patients with ACC [9-11].
Despite growing interest, immunotherapy has shown limited benefit in ACC. The Tumor microenvironment (TME) of ACC is typically characterized by an “immune desert” phenotype, with markedly reduced infiltration of effector immune cells such as CD8+ T cells and dendritic cells compared to immunologically “hot” tumors like mela- noma and lung cancer, which hinders the initiation and effect of anti-tumor immune response [12, 13]. Several targeted and immune-based therapies, including lenvatinib, cabozantinib, and pembrolizumab, have been investigated [14, 15]. Among them, Len- vatinib has demonstrated a disease control rate of approximately 60% [16, 17], but the duration of response is short, and there is an absence of reliable biomarkers for pre- dicting treatment efficacy [18]. Given the rarity of ACC and the challenges of single- modality approaches, integrating multi-omics profiling with advanced machine learning provides an opportunity to uncover new therapeutic targets and predictive biomarkers [19-21].
Previous efforts at ACC classification have predominantly relied on transcriptomic clustering or single-omics analyses, which yielded prognostic insights but limited ther- apeutic implications [22]. To address this gap, we conducted an integrative analysis of transcriptomic, epigenetic, and genomic variation data to refine ACC molecular sub- types and assess their clinical relevance. By applying multiple clustering algorithms and validating with cluster prediction index (CPI) and Gap statistics (a statistical method for determining the number of clusters), we identified two robust subtypes with reproduc- ible biological distinctions across omics layers [23]. These subtypes exhibited differen- tial proliferative activity, immune contexture, and metabolic states, providing clinically meaningful insights into tumor behavior and treatment response. On this basis, we propose the Multi-Omics ACC Consensus Subtyping (MACCS) model, which bridges the gap between prognostic classification and actionable clinical decision-making by linking molecular subtype identity with both biological mechanisms and therapeutic vulnerabilities.
2 Materials and methods
2.1 Dataset collection
Multi-omics data of ACC patients were obtained from The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/) [24], including transcriptome profiles, DNA methylation, somatic mutations, and corresponding survival information. Only patients with complete multi-omics and prognostic data were included in the analysis. For exter- nal validation, four independent ACC gene expression datasets were retrieved from the Gene Expression Omnibus (GEO) database: GSE10927 (n=24), GSE33371 (n=23), GSE70621 (n=29), and GSE19750 (n=22). As all data were derived from public reposi- tories, no additional ethics approval was required.
2.2 Identification and validation of multi-omics molecular typing
To ensure stable and reliable molecular subtyping, we applied a multi-omics consensus analysis strategy. Prognosis-associated features were first identified within each omics layer using univariate Cox regression. These features were then input into the R pack- age MOVICS (version 1.0) [25], which integrates multiple clustering algorithms and provides outputs on subtype characteristics, prognostic differences, and therapeutic sensitivity. We used 10 clustering algorithms from MOVICS (COCA, Consensus Clus- tering, CLML, IntNMF, iClusterBayes, PINSPlus, NEMO, MoCluster, LRAcluster, and SNF) to cluster prognostic features. We perform consensus analysis on the clustering results of each algorithm using combined classification of the consensus set to obtain clustering results with high robustness.
The optimal number of clusters was determined by combining prior biological knowl- edge with two quantitative indices: the Cluster Prediction Index (CPI) and Gap statistics [26, 27]. The number that resulted in the maximum value of gap statistics and CPI was selected as the optimal number of clusters for the input data. To evaluate the robustness and reproducibility of the multi-omic subtypes identified from the TCGA-ACCs dataset, we performed nearest template prediction (NTP, a supervised classification approach that assigns samples to predefined classes based on correlation to class templates) analy- sis using the runNTP function in the MOVICS package.
2.3 GO, KEGG, and GSVA analysis
In this study, the DESeq2 software package [28] was used to calculate the differentially expressed genes (DEGs) between MACCS1 and MACCS2, with the application parame- ters of |log2FC| >1 and p < 0.05. To better understand the biological differences between the two groups, the DEGs were subjected to Gene Ontology (GO) and Kyoto Encyclope- dia of Genes and Genomes (KEGG) pathway functional enrichment analysis using the clusterProfiler package (version 4.14.6) in R language. GO analysis included three clas- sification levels: cellular component (CC), biological process (BP), and molecular func- tion (MF), and the top 10 terms in each category were displayed to analyze the biological characteristics of each molecular subtype. KEGG pathways can be used to identify potential signaling pathways. We used the GSVA package to perform single-sample gene set enrichment analysis (ssGSEA) [29], and the limma software package [30] to analyze the differences in pathway scores between subgroups. The results of all biological path- way enrichment analyses were visualized using the ggplot2 toolkit (version 3.4.0), and a p < 0.05 value for enriched words was considered significantly enriched. The gene sets included in our analysis are all from the built-in gene sets in MSigDB, IOBR, and MOV- ICS software packages.
2.4 Immune infiltration analysis and genomic analysis
To investigate the differences in responses to immunotherapy between the two groups, we used multiple algorithms to compare the differences in immune components between the two groups, including immune cells, expression levels of immune checkpoint-related molecules, and immune scores. We studied the infiltration of individual immune cells in the tumor by ssGSEA and calculated the built-in immune-related gene scores of IOBR and MOVICS in each sample; The maftools package [31] was used to calculate the TMB
of each sample and visualize the mutation map. Finally, the sensitivity of different sub- groups to immunotherapy was compared by using the TIDE algorithm [32].
2.5 Drug sensitivity analysis
This study analyzed the sensitivity of different subgroups of patients to various drugs by using the drug sensitivity data of tumor cell lines in the Genomics of Drug Sensitiv- ity in Cancer (GDSC, a public resource linking genetic features of cancer cell lines to drug response profiles) database (https://www.cancerrxgene.org/) [33]. The half-maxim al inhibitory concentration (IC50) of each sample was calculated using pRRophetic in R language. The lower the IC50 value, the more sensitive the sample is to the drug. In addi- tion, we calculated potential treatment-sensitive targets in poor prognosis subgroups based on the Connectivity Map (cMap) database [34].
2.6 In vitro experiment
The ACC cell line NCI-H295R [35] used in the in vitro experiments of this study was obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). The cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin at 37 ℃ and 5% humidity. Cells were routinely passaged every 2-3 days, and all experiments were performed dur- ing the logarithmic growth phase. NCI-H295R cells were selected because they retain multiple steroidogenic enzyme activities across adrenal cortex zones, aligning them with MACCS1’s proliferative and steroidogenic features [36]. The effect of HOXC11 interfer- ence on ACC cell proliferation was verified using a cell counting kit, and the cell experi- ment was evaluated according to the instructions, and the absorbance was measured at a wavelength of 450 nm using a microplate reader.
2.7 Statistical test
All data processing, statistical tests, and visualizations in this study were performed using R language (version 4.5.0) and SPSS software (version 29.0). The Wilcoxon rank sum test or t test was used to compare the differences in continuous variables between the two groups. Cox regression and Kaplan-Meier analysis were used for survival analy- sis. A two-tailed test was used, and a P value <0.05 was set as statistically significant. No specific code was generated in this study, and all analyses were completed using the default parameters of the corresponding software and R package.
3 Results
3.1 Identification of novel ACC molecular subtypes based on multi-omics consensus clustering
To systematically characterize the molecular heterogeneity of ACC and construct a refined molecular classification system, we integrated transcriptomic (mRNA, lncRNA, miRNA), DNA methylation, and somatic mutation data from the TCGA and GEO data- sets. Using ten multi-omics clustering algorithms implemented in the MOVICS frame- work, we performed unsupervised consensus clustering. Both the CPI and Gap Statistics consistently indicated that 2 clusters represented the optimal solution (Fig. 1A). The concordance between the consensus matrix and the subtype profiles further confirmed the robustness of the classification, leading us to define two novel molecular subtypes,
A
B
C
1.0
1.0
Cluster Prediction Index
Subtype
0.8
0.8
Matrix
1
Gap-statistics
0.8
0.6
0.6
0.6
0.4
0.2
0
1: 40 | 0.43
0.4
0.4
Subtype
MACCS1
MACCS2
0.2
0.2
0.0
0.0
2: 39 | 0.45
2
3
4
5
6
7
Number of Multi-Omics Clusters
8
-
€
D
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
Integrated Multi-Omics Profiles of ACC
AJCC
Silhouette width
I
E
Subtype - MACCS1 - MACCS2
AJCC
II
1.00
pstage
III
age
IV
Overall survival
0.75
Subtype
X
0.50
- MT-ATP8 MT-ND4
MT-ND4L
pstage
0.25
Log-rank
MT-ND5
p< 0.0001
MT-CO2
T1
MT-CO1
MT-CYB
0.00
mRNA
T2
- STAR
0
2.5
5
7.5
10
12.5
RPL13A
T3
Time (years)
- TMSB10
T4
Number at risk
cluster
age
40
21
5
3
2
1
80
39
29
19
7
2
1
0
2.5
Time (years)
5
7.5
10
12.5
60
40
Progression Free Survival
1.00
20
0.75
S
- SNHG19
IncRNA
DANCR
0
- MIR202HG
Subtype
0.50
G
MACCS1
0.25
- GASS
Log-rank
ZFAS1 SNHG6
MACCS2
0.00
p < 0.0001
0
2
4
6
8
10
12
LINCO0998 INC00086
mRNA
Time (years)
LINC00087
2
Number at risk
RGS5
1
cluster
40
8
2
1
1
1
1
hsa-mir-150
hsa-mir-125b-1
0
39
33
18
11
5
2
1
0
2
4
6
Time (years)
3
10
12
-1
miRNA
1.00
hsa-mir-182
-2
Disease Specific Survival
hsa-mir-183
hsa-mir-139
IncRNA
0.75
hsa-mir-198a-1
hsa-mir-196a-2
2
hsa-mir-198b
- hsa-mir-153-2
1
0.50
- hsa-mir-1269a
0
0 25
Log-rank p < 0.0001
og01724917
-1
0.00
- cg14391855
-2
0
2
4
6
8
- cg11776014
Time (years)
10
12
Methylation
miRNA
Number at risk
- 0900250430
2
cluster
39
22
7
4
2
2
1
cg25143990
1
38
35
22
12
6
2
1
cg02599464
0
2
4
6
B
10
12
cg03816707
0
Time (years)
-0902746725
-1
Disease Free Interval
1.00
9
cg08927006
cg02630214
-2
0.75
11
M value
.
..
~ TTN
.
.
LRP1
2
0.50
.
- FAT1
1
0.25
Log-rank
.
.
.
.
CNTNAP5
p =0.00056
Mutation
SPTA1
.
.
- DST
- NF1
0
0.00
-1
0
2
4
6
Time (years)
8
10
12
#
.
.
- SVEP1
-2
Number at risk
M
Mutated
cluster
12
7
2
1
1
1
1
.
- PKHD1
A
- TP53
0
33
28
16
10
5
2
1
A
M
1
11
.
0
2
4
Time (years)
6
B
10
12
1
designated MACCS1 and MACCS2 (multi-omics based ACC subtype 1 and 2; Fig. 1B, C).
We next examined subtype-specific molecular features across omics Layers. The top 10 differential features from each layer were integrated into a comprehensive heatmap (Fig. 1D). At the mRNA level, MACCS1 tumors showed elevated expression of SNHG19, MIR202HG, and GAS5, genes implicated in cell migration and immune evasion, and associated with unfavorable prognosis. By contrast, MACCS2 tumors exhibited
enrichment of mitochondrial genes such as MT-ATP8, MT-ND4, and MT-ND4L, sug- gesting enhanced oxidative phosphorylation, metabolic reprogramming, and potential resistance to stress or metastatic adaptation. At the miRNA level, MACCS1 tumors were enriched for miR-183, miR-139, and miR-196b, which likely exert post-transcrip- tional regulation of oncogenic pathways by targeting cell cycle and mRNA stability. DNA methylation analysis revealed subtype-specific CpG modifications (cg01724917 and cg11776014), indicating an important role for epigenetic mechanisms in subtype formation. Mutational profiling showed that MACCS1 harbored higher mutation fre- quencies in key tumor suppressors such as TP53 and CNTNAP5, consistent with greater genomic instability and disease aggressiveness. Clinically, survival analysis demonstrated that patients with MACCS1 had significantly worse overall survival (OS), progression- free survival (PFS), disease-specific survival (DSS), and disease-free interval (DFI) com- pared with MACCS2 (Fig. 1E), underscoring MACCS1 as a high-risk subtype with poor prognosis.
To validate the robustness and reproducibility of this classification, we performed NTP algorithm across four independent GEO datasets (GSE10927, GSE33371, GSE7062, and GSE19750). In each dataset, patients were reliably assigned to MACCS1 or MACCS2, and the heatmaps of expression patterns closely recapitulated the subtype signa- tures observed in TCGA (Fig. 2A and D). Survival analysis showed that patients with MACCS1 subtype had markedly inferior outcomes compared with those with MACCS2 subtype. Collectively, these findings demonstrate that MACCS1 and MACCS2 represent reproducible and clinically meaningful ACC subtypes, highlighting their potential utility for patient stratification and prognostic assessment.
3.2 MACCS classification is associated with proliferation behavior in ACC
To explore biological differences between the MACCS subtypes, we conducted differen- tial expression and enrichment analyses in the TCGA-ACC cohort. GO analysis showed that in the biological processes (BP) category, genes such as APOA1 and ENTPD8 were upregulated in MACCS1, suggesting a role for metabolic reprogramming in fueling tumor invasion and metastasis. Elevated expression of genes including HOXD13 and OTX1 further indicated that MACCS1 may promote neuron-like migratory proper- ties, thereby facilitating neural and distant metastasis (Fig. 3A). Functional annotation revealed that differentially expressed genes in MACCS1 were enriched in cell-cycle related pathways such as chromatid segregation, DNA replication, and organelle divi- sion (Fig. 3B). Consistently, MOVICS-based gene set analysis highlighted enrichment in developmental and differentiation processes, including tissue regionalization, aligning with the aggressive phenotype of this subtype (Fig. 3C). GSVA analysis confirmed that MACCS1 was associated with activation of canonical oncogenic pathways, including E2F targets, G2/M checkpoint, and PI3K/AKT/mTOR signaling. In contrast, MACCS2 was enriched in tumor-suppressive and immune-associated pathways such as KRAS sig- naling, interferon-y response, and angiogenesis (Fig. 3D), underscoring the functional divergence between the two subtypes. KEGG pathway analysis further demonstrated that MACCS1-specific genes were concentrated in pathways linked to complement and coagulation cascades and drug metabolism, suggesting involvement in immune regula- tion, and metabolic homeostasis (Fig. 3E).
A
Validation in GSE10927
100
MACCS1
MACCS2
Survival probability (%)
75
template features
50
P=0.002
25
0
0
12
24
36
48
60
72
84
96
Time (Months)
108
120
Number at risk
11
3
1
1
1
1
1
1
1
0
0
p value
MACCS1
MACCS2
O-
13
12
9
4
4
2
1
1
1
1
1
class predictions
0
12
24
36
48
60
72
Time (Months)
84
96
108
120
B
Validation in GSE33371
100
MACCS1
MACCS2
75
template features
Survival probability (%)
50
P=0.001
25
0
0
12
24
36
48
60
72
Time (Months)
84
96
108
120
Number at risk
p value
b
11
3
1
1
1
1
1
1
1
1
1
MACCS1
MACCS2
12
11
8
6
6
5
4
3
2
1
class predictions
1
0
12
24
36
48
60
72
84
96
108
120
C
Validation in GSE70621
Time (Months)
100
MACCS1
MACCS2
75
template features
Survival probability (%)
50
P=0.004
25
0
0
12
24
36
48
60
Time (Months)
72
84
96
108
120
Number at risk
14
10
5
3
2
2
2
2
2
1
1
p value
0-
15
13
11
10
10
10
8
8
8
6
4
MACSS1
class predictions
MACCS2
0
12
24
36
48
60
72
84
96
120
D
Validation in GSE19750
Time (Months)
108
100
MACCS1
MACCS2
Survival probability (%)
75
template features
50
P= 0.094
25
0
0
12
24
36
48
60
72
84
96
108
120
Time (Months)
Number at risk
O-
10
7
6
2
1
1
1
1
1
1
1
p value
MACCS1
MACCS2
0
12
10
9
8
8
8
7
6
5
4
3
class predictions
0
12
24
36
48
60
72
84
96
108
Time (Months)
120
A
NS
Log2 FC
p-value
p-value and log2 FC
B
Up regulated in MACCS1
mitotic sister chromatid
segregation
chromosomal region
30
APOA1
sister chromatid segregation
OTX1
-Log 10 P
ENTPD8
MUČL1
mitotic spindle organization
p.adjust
SEPT5
20
CEx2 FORD1
MYH4
C1QTNF8
SUPI
PLCL
chromosome, centromeric region
COOPOK
KD1
TNR
P73 HOND13
Sort223
6.716667e-08
PRŠS1 UGT 1A7
PEBP4 PIFO
RIFZC
COCZO HAIPL? ·
SORCS1 1SL1
EZF1
chromosome segregation
FAM25A
OFMOSIQUE
10
MYBPI
25
COHID . FOKAZ
H83514 RGSBLAN
SP9
spindle organization
KCNC2
TECAL GJ
4
NKX2-5
MUC15
Cfort
DNA replication
9817
SRY
NHL12
83A
0
KLHOCT
nuclear division
-5
0
5
10
organelle fission
Log2 fold change
0
2
4
6
C
GO_REGIONALIZATION
GO_NEURON_FATE_COMMITMENT
Subtype
GO_PATTERN_SPECIFICATION_PROCESS
GO_EMBRYONIC_SKELETAL_SYSTEM_DEVELOPMENT
MACCS1
GO_PROXIMAL_DISTAL_PATTERN_FORMATION
MACCS2
GO_DORSAL_VENTRAL_PATTERN_FORMATION
GOTANTERIOR_POSTERIOR_PATTERN_SPECIFICATION
GO NEURON_FATE_SPECIFICATION
NES
GO_SISTER_CHROMATID_SEGREGATION
1
GO_CELL_DIFFERENTIATION_IN_SPINAL_CORD
GO_COMPLEMENT_ACTIVATION
0.5
GO_RETINOL_METABOLIC_PROCESS
GOTANTIGEN_PROCESSING_AND_PRESENTATION_OF_ENDOGENOUS_ANTIGEN
0
GO_URONIC_ACID_METABOLIC_PROCESS
-0.5
GO_CELLULAR_GLUCURONIDATION
-1
GO_NEGATIVE_REGULATION_OF_EXECUTION_PHASE_OF_APOPTOSIS
GO_RESPIRATORY_ELECTRON_TRANSPORT_CHAIN
GO_NEGATIVE_REGULATION_OF_SIGNALING_RECEPTOR_ACTIVITY
GO_PRIMARY_ALCOHOL_METABOLIC_PROCESS
GO_POSITIVE_REGULATION_OF_LEUKOCYTE_MEDIATED_CYTOTOXICITY
D
G2M_CHECKPOINT
E
MYC_TARGETS_V1
DNA_REPAIR
E2F_TARGETS
KEGG enrichment barplot
UNFOLDED_PROTEIN_RESPONSE
MYC_TARGETS_V2
Neuroactive ligand-receptor interaction
TGF_BETA_SIGNALING
MTORCI_SIGNALING
Drug metabolism - cytochrome P450-
WNT_BETA_CATENIN_SIGNALING
-log10(pvalue)
MITOTIC_SPINDLE
Metabolism of xenobiotics by cytochrome P450-
PI3K_AKT_MTOR_SIGNALING
UV_RESPONSE_UP
Retinol metabolism -
9
OXIDATIVE_PHOSPHORYLATION
P53_PATHWAY
Drug metabolism - other enzymes-
7
PROTEIN_SECRETION
ADIP OGENESIS
Complement and coagulation cascades-
TNFA_SIGNALING_VIA_NFKB
5
ANDROGEN_RESPONSE
Pentose and glucuronate interconversions-
SPERMATOGENESIS
3
NOTCH_SIGNALING
Protein digestion and absorption -
PEROXISOME
JEME_METABOLISM
Tyrosine metabolism
Activated in MACCS1
L6 JAK STAT3_SIGNALING
REACTIVE_OXIGEN_SPECIES_PATHWAY
Chemical carcinogenesis - DNA adducts-
Not Significant
APICAL SURFACE
CHOLESTEROL_HOMEOSTASIS
cAMP signaling pathway
Activated in MACCS2
APICAL JUNCTION
EPITHELIAL_MESENCHYMAL_TRANSITION
Chemical carcinogenesis - receptor activation-
UV_RESPONSE_ON
HEDGEHOG_SIGNALING
ECM-receptor interaction -
COAGULATION
APOPTOSIS
Bile secretion
FATTY_ACID_METABOLISM
GLYCOLYSIS
Pancreatic secretion
HYPOXIA
IL2_STAT5_SIGNALING
Calcium signaling pathway
PANCREAS_BETA_CELLS
ALLOGRAFT_REJECTION
Cytokine-cytokine receptor interaction-
XENOBIOTIC_METABOLISM
ESTROGEN_RESPONSE_EARLY
Cholinergic synapse
COMPLEMENT
KRAS_SIGNALING_DN
Viral protein interaction with cytokine and cytokine receptor-
KRAS_SIGNALING_UP
ESTROGEN_RESPONSE_LATE
Steroid hormone biosynthesis
INTERFERON_ALPHA_RESPONSE
INFLAMMATORY_RESPONSE MYOGENESIS
0
10
20
30
40
Number of Gene
ANGIOGENESIS
INTERFERON_GAMMA_RESPONSE
BILE ACID METABOLISM
-5
0
5
+
Activated in MACCS1
t value of GSVA score, MACCS1 versus MACCS2
To validate these phenotypic differences, we further annotated MACCS2 with IOBR- derived gene sets. Metabolic pathways, including homocysteine biosynthesis, methio- nine cycle, purine/pyrimidine metabolism, and folate one-carbon metabolism, were significantly downregulated (Figure S1A). In contrast, immune-related pathways, includ- ing immune cell proliferation and infiltration signatures, were markedly upregulated, consistent with MACCS2 representing an “immune-hot” phenotype associated with more favorable prognosis (Figure S1B).
3.3 Immune landscape profiling reveals MACCS2 as an immune-hot, checkpoint- responsive subtype in ACC
To investigate the immunological basis underlying differential therapeutic sensitivity of the MACCS subtypes, we compared immune-related gene expression, immune cell infiltration, and predictive markers of immunotherapy response between MACCS1 and MACCS2. Key immune checkpoint molecules, including CD274 (PD-L1), CD247, and PDCD1LG2 (PD-L2), were significantly upregulated in MACCS2, suggesting heightened sensitivity to immune checkpoint inhibitors (ICIs). Consistently, infiltration of mul- tiple immune cell populations including activated CD4 memory T cells was markedly higher in MACCS2, forming an “immune-hot” TME (Fig. 4A). ESTIMATE analysis fur- ther confirmed significantly higher immune, stromal, and composite scores in MACCS2
A
ImmuneScore
ImmuneScore
B
Subtype
MACSCS1
MACCS2
2
7
StromalScore
Subtype
1
CD274
0
1
:
PDCD1
Signature_score
CV
CD247
2
PDCD1LG2
StromalScore
CTLA4
2
TNFRSF9
1
0
TNFRSF4
0
T.cells.CD8
1
II
T.cells.regulatory .. Tregs.
-2
StromalScore
ImmuneScore
ESTIMATEScore
T.cells.CD4.naive
T.cells.follicular.helper
Subtype
B.cells.naive
MACCS1
MACCS2
C
B.cells.memory
T.cells.gamma.delta
2
Dendritic.cells.activated
ICI
wilcoxtest p = 6.5e-09
Macrophages. M1
2
NK.cells.activated
1
(TMB + 1)
Plasma.cells
0
T.cells.CD4.memory.resting
1
T.cells.CD4.memory.activated
Mast.cells.activated
-2
NK.cells.resting
TIME
0
MACCS1
MACCS2
Macrophages. MO
1
Macrophages. M2
0.5
…
Eosinophils
0
Monocytes
-0.5
D
Dendritic.cells.resting
-1
100%
P=0.01
Mast.cells.resting
0%
Response
90%-
Neutrophils
Me TIL
33%
Endothelial cells
1
80%-
True
70%-
0.5
False
Fibroblasts
60%-
0
50%.
0.5
40%-
92%
MeTIL
30%-
67%
-1
20%-
10%-
0% -
E
MACCS1
MACCS2
Subtype
MACSCS1
MACCS2
ns
*
ns
”
Signature_score
-
T
Y
CD_8_T_effector
Immune_Checkpoint
APM
TMEscoreA
Mismatch_Repair
Nucleotide_excision_repair
DOR
DNA_replication
Base_excision_repair
Pan_F_TBRS
EMT1
EMT2
EMT3
TMEscoreB
compared with MACCS1 (Fig. 4B). Interestingly, although MACCS1 displayed higher tumor mutational burden (TMB) and was theoretically more immunogenic (Fig. 4C), it lacked substantial immune activation, indicating the likely presence of immune evasion mechanisms.
To assess predicted response to immunotherapy, we applied the TIDE algorithm. Results indicated that MACCS2 patients were more likely to respond to ICIs, consistent with their higher expression of immune checkpoint genes (Fig. 4D). Pathway enrichment analysis revealed that MACCS2 was strongly associated with enhanced CD8+ T cell activity, antigen presentation, and TME activation, whereas MACCS1 showed enrich- ment in pathways related to epithelial-mesenchymal transition (EMT) and DNA repair (MMR, NER, BER), which may promote immune escape by stabilizing the genome and reinforcing immunosuppressive signaling (Fig. 4E).
Collectively, these findings identify MACCS2 as an immune-hot subtype with a favor- able immunotherapy profile, suggesting that ICI-based regimens may represent an effec- tive strategy for this group. By contrast, the immunosuppressive phenotype of MACCS1, despite its high TMB, may render it more suitable for targeted agents or rational combi- nation therapies.
3.4 Subtype-specific drug sensitivity profiling identifies candidate therapeutics for MACCS1 and MACCS2 in ACC
To explore therapeutic vulnerabilities of the ACC molecular subtypes, we compared predicted drug sensitivities between MACCS1 and MACCS2. Using the MOVICS pack- age, we found that MACCS1 tumors exhibited greater sensitivity to the tyrosine kinase inhibitors axitinib and pazopanib (Fig. 5A). Integration with CMAP database analysis further identified candidate small molecules with potential efficacy in MACCS1, includ- ing arachidonoyl trifluoromethane, butein, imatinib, NU-1025, and copper sulfate
A
C
4.0-
·
Gemcitabine
Epothilone.B
TW37
GSK.650394
Embelin
1.
1.0.
4.8-
65-
Estimated ICso of Axitinib
Estimated IC50
2-
Estimated IC50
Estimated IC50
0.5
Estimated IC50
Estimated IC50
6
55.
*
·
e
0
4
-10-
-4-
MACCS1
MACCS2
MACCS1
MACCSZ
MACCS1
FA, MACCSZ
4.0=
MACOS1
MACCSZ
35-
MACCS1
MACCSZ
25
Micaxon, p = 2.48-11
Sorafenib
Vinorelbine
3
MACCSS
MACCS2
GWB43582X
S. Trityl.L.cysteine
49-
OD
20-
$0.
4-
Estimated ICsg of Pazopanib
Estimated IC60
4.8-
4.7-
Estimated IC50
Estimated IC50
Estimated IC60
-1-
Estimated IC50
14-
2
5.
18
·
18-
2
0
8.
15-
0
D
4
-8-
0.5.
&G
·
4.3-
MACCST
sefifi MACCS2
-2-
NACCSZ
65.
·
MACCSI
MACCS1
SVAMACCS2
MACOS1
Lenalido MACCSZ
DO.
MACCS1
ACCS2
0%
I
14
D
Gefitinib
LFM.A13
Lenalidomide
JNJ.26854165
·
·
35.
Micoion. pm 5. 5e-05
30-
7.00
83-
MACCS1
MACCS2
Estimated IC50
Estimated IC50
Estimated IC50
4.50-
Estimated IC50
Estimated ICS0
B
A
25
1.0
1
8.50-
15
4.00-
CMAP score
·
·
0.5
MACCS1
CCTOO MACCS2
6.8-
MACCS1
MACCSZ
MACOS1
MACOS1
MACCS2
20-
05-MACCS2
MACCS1
AACCS2
CCT007093
CI.1040
AZD.0530
VX.702
AZD6244
0.0
NU. 1025
copper.sulfate
35.
butein
imatinib
Estimated IC50
5.50-
5.0-
3.
Estimated IC50
5.50-
Estimated IC50
Estimated IC50
Estimated IC50
3D-
25-
-0.5
25-
5
100
-
100
-1.0
·
arachidonyitrifluoromethane
.
*
15-
MACCS1
MACCS2
3.5
.
MACCS1
MACCISZ
q-
MACCS1
MACCS2
MACOS1
MACCSZ
MACCS1
MACCS2
0
25
50
75
100
Rank number
(Fig. 5B). These compounds are predicted to exert antitumor effects through inhibi- tion of the AKT/mTOR pathway and induction of apoptosis or oxidative stress-related autophagy.
We next applied the GDSC resource to predict IC50 values across subtypes. In MACCS1, predicted sensitivity was observed for multiple agents such as gem- citabine, epothilone-B, TW-37, GSK-650,394, embelin, sorafenib, vinorelbine, IPA-3, GW843682X, and S-trityl-L-cysteine, supporting this subtype’s susceptibility to cyto- toxic drugs and kinase inhibitors (Fig. 5C). Conversely, MACCS2 displayed higher pre- dicted sensitivity to a distinct set of targeted compounds, including gefitinib, LFM-A13, GNF-2, lenalidomide, JNJ-26,854,165, CCT007093, CI-1040, AZD-0530, VX-702, and AZD6244 (Fig. 5D). Most of these agents act as inhibitors of the EGFR, MAPK, and JAK/STAT pathways, suggesting that MACCS2 tumors may be particularly dependent on these signaling cascades.
3.5 MACCS1 and MACCS2 exhibit distinct genomic profiles
To delineate the genomic characteristics of the MACCS subtypes, we compared their mutational spectra, co-mutation networks, and copy number variation (CNV) patterns. In MACCS1, the five most frequently mutated genes were CTNNB1 (17%), TP53 (17%), TTN (13%), MUC16 (11%), and CNTNAP5 (10%) (Fig. 6A). By contrast, the mutational profile of MACCS2 was dominated by MUC16 (36%), followed by HMCN1 (21%), NLN (21%), BOP1 (14%), and CMYA5 (14%) (Fig. 6A). Most of these alterations involve struc- tural or metabolic genes and are likely “passenger” mutations with limited translational relevance, except for BOP1, which has been implicated in oncogenic processes. Notably, TP53, CTNNB1, and TTN mutations were enriched in MACCS1 and correlated with higher AJCC and pathological stages (Fig. 6B).
Co-mutation analysis revealed that MACCS1 exhibited a more complex mutational coordination pattern, with significant co-occurring pairs such as MYO15A-DST, MYO15A-FNBP3, and MYO15A-FRAS1, implying dysregulation of highly intercon- nected signaling pathways. In contrast, MACCS2 showed fewer significant co-mutation relationships (Fig. 6C). Analysis of CNVs showed no significant difference in the over- all burden of genomic alterations (FGA) between subtypes. Likewise, the frequencies of copy number gains (FGG) and losses (FGL) were comparable between MACCS1 and MACCS2, indicating that CNV is unlikely to be the major determinant of subtype diver- gence (Fig. 6D).
In summary, MACCS1 is characterized by enrichment of canonical driver muta- tions and complex co-mutation networks consistent with genomic instability, whereas MACCS2 harbors a higher frequency of largely non-driver mutations with relatively mild genomic disruption.
3.6 HOXC11 is a potential diagnostic and therapeutic target for ACC patients
Given the markedly poor prognosis of the MACCS1 subtype, we applied random forest and univariate Cox regression analyses to prioritize subtype-specific prognostic genes. HOXC11 emerged as the top-ranked candidate with the highest prognostic impor- tance score, identifying it as a key molecular marker of MACCS1 (Fig. 7A). In addition, across multiple datasets, high HOXC11 expression was significantly associated with worse OS, DFS, DSS, and PFS (HR>1, P<0.05, Fig. 7B). Moreover, HOXC11 expression
A
B
#
MACCS1
MACCS1
M
,
1
16%
TP53
a .
U
a
Min di sản piva
11
0 .
0
$
CINNB
17 %
M/C16
36%
TP53
17%
HMCN 1
21%
TIN
13%
MACHS
AN
21%
15%
CTNNB1
11%
BOP1
14%
CN THAPS
10%
CAIXAS
14%
PRI-01
II
10%
CA2
14%
APOB MEN’S
8%
FBN2
14%
8%
GL/2
14%
8%
AMT20
1
11%
TTN
SVEP1
I
MYLA2
14%
ASYLS
6%
PCDH8 12
14%
DSr
6%
PLIN83
14%
FAN
6%
TNR
14%
AJCC
FBNS
6%
USP34
14%
I
pstage
FRAS1
6%
APHGAP44
7%
FROMPOR
6%
CACMAMA
1
age
GNAS
6%
FRMDO
Subtype
GOLGAT
FSTL4
%
AM728
POCO11
Te
7%
AJCC
pstage
age
Subtype
LAP1
6%
SLC25437
7%
!
T1
80
60
MACCS1
T2
MACCS2
Missense_Mutation
Nonsense_Mutation
m
T3
IN
40
T4
20
Frame_Shift_Del
Splice_Site
x
0
In_Frame_Del
Multi_Hil
Frame_Shift_Ins
C
CTNNB1
CNTNAP5
TTN
MUC16
ADAMTS17
MACCS1
TP53
PKHD1
APOB
MENT
FRMPD4
GOLGA4
AC006059 2
ARHGAP44
SLC25A37
NF1
SVEP1
ASXL3
DST
FAT 4
FBN3
FRAS1
GNAS
ITIHİ
KMT2B
LRP1
PRKARTA
TUTT
MYO15A
MUC16
HMCN1
BOP1
CMYA5
FBN2
KMT2D
MYLK2
PCDHB12
PLXNB3
USP34
ABCD2
ABLIM3
CACNA1A
FRMD8
FSTL4
PDCD11
TTC16
UNC5D
MACCS1
NLN
CR2
GLI2
TNR
MYO15A
.
*
*
*
1
UNCSD
ADAMTS17
.
.
*
.
*
*
+
*
*
TTC16
TUT7
SLC25437
PRKARTA
PDCD11
LRP1
*
*
.
+
-
FSTL4
KM728
.
.
K
.
.
FRMD8
ITİH1
.
CACNA1A
GOLGA4
.
.
GNAS
.
+
-
.
.
.
*P<40
ARHGAP44
AC006059.2
*P<00
.
FRMPD4
.
*
.
.
.
. P<00
ABLIM3
· P<005
FRAS1
.
.
ABCD2
FBN3
.
.
USP34
FAT4
.
.
.
TNR
.
DST
.
*
*
.
PLXNB3
ASXL3
.
PCDHB12
.
.
SVEPİ
.
MYLK2
NF1
.
.
*(Co-occurina)
KMT2D
.
3 (Co-connc)
MEN1
GU2
.
.
APOB
2
FBN2
2
PKHD1
1
CR2
.
1
CNTNAP5
-logto(-vale)
CMYA5
MUC16
0
BOP1
0
TIN
1
NLN
1
TP53
HMCN1
CTNNB1
2
MUC16
2
> 3(Adsaly ecknive)
·3(Mutualy archive)
D
Copy number-altered genome
Copy number-lost genome
Copy number-gained genome
MACCS2
.
.
.
MACCS1
0.6
0.4
0.2
0.0
-0.2
0.0
0.2
FGA (Fraction of Genome Altered)
FGL or FGG (Fraction of Genome Lost or Gained)
was markedly elevated in ACC tumor tissues compared with adjacent normal tissues (Fig. 7C), and was significantly higher in advanced-stage tumors (T3/T4) than in early- stage tumors (T1/T2) (Fig. 7D), implicating its role in disease progression.
To investigate the genomic basis of aberrant HOXC11 expression, we stratified TCGA- ACC samples into high- and low-expression groups using the median HOXC11 level as the cutoff. The high-expression group displayed increased mutation frequencies in TP53 and PKHD1, along with higher frequencies of deletion events at 9p23 and 9p21.3, sug- gesting disruption of tumor suppressor pathways (Fig. 7E). Over-representation analysis
A
B
a
HOXC11
Cox regression anlaysis
ISL2
ISL1
OS_TCGA_ACC
COL11A1
SV2C
OS_GSE10927
.
ASB17
OS_GSE33371.
GREM2
APOA1
OS_GSE19750-
Error Rate
POU4F1
TBX4
DFS_TCGA_ACC
**
:30
COL2A1
DFS_GSE76019
FOXA2
ENTPD8
DFS_GSE76021
CDHR5
SORCS1
DSS_TCGA_ACC.
LMX1B
PFS_TCGA_ACC
:
CCDC83
C6orf223
0.5
1.0
3.0
FREM2
Hazard ratio
RP11_35NB.
HOXD13
OS
DFS
SS
PFS
·
200
600
800
1000
082
Number of Trees
0.04
C
Variable Importance
D
GSE10927
GSE12368
GSE143383
GSE33371
GSE90713
TCGA_ACC
Wilcoxon, p = 0.02
Willcoxon, p = 0.033
4
Wilcoxon, p = 0.019
Wilcoxon, p = 0.02
Wilcoxon, p = 0.028
3
Kruskal-Wallis, p = 0.0016
HOXC 11 Expression (z-score)
HOXC 11 Expression (z-score)
3
HOXC 11 Expression (z-score)
3
HOXC 11 Expression (z-score)
HOXC 11 Expression (z-score)
HOXC 11 Expression (z-score)
2
2
2
2
1
1
2
1
O
D
1
.
0.
0
0
-1
-1
=
Normal
Empor
Norte
Error
Normal
BUamor
Normal
Amor
Normna
Lamos
5
”
-
4ª
E
Low HOXC11
High HOXC11
Percent
F
GSEA-Hallmark Analysis
TP53
-
¥
CINNB1
G2m checkpoint
MUCHE
E 2f targets
AMONT
TIM
Myc targets v1
PKHD1
Mitotic spindle
CNTNAPS
Dna repair
FAT4
Unfolded protein response
MENT
Wnt beta catenin signaling
ARIC4
APOB
Myc targets v2
DST
PI3k akt mtor signaling
HRNIR
Tgf beta signaling
p.adjust
KM728
OBSGN
Glycolysis
0.75
PODH15
Gun
Notch signaling
0.50
PRKARTA
Uv response up
0.25
STABI
Spermatogenesis
SVEPI
ZCCHC6
Mtorc1 signaling
5015.33
Pancreas beta cells
5g35.3
Apical surface
tp36.23
Cholesterol homeostasis
3013.31
Apical junction
4935.1
Sp23
Kras signaling dn
$=21.3
Peroxisome
13q14.2
Hedgehog signaling
17q21.2
Apoptosis
17q24.2
22q12.1
0.0
0.2
Enrichment Score
0.4
0.6
ã
8
G
4.
H
si_NC
si_NC
si_HOXC11
100
OD values/450nm
**
si_HOXC11
Number of colonies/well
3
75-
2
50
a
1
25
O
0
0
24
48
72
0
Hours
Si_NC
si_HOXC11
further demonstrated enrichment of cell cycle-related signaling pathways, including the G2/M checkpoint and E2F targets, in the HOXC11-high group (Fig. 7F), supporting its role as a driver of proliferative signaling in ACC.
Functional validation was performed in vitro. Silencing HOXC11 expression signifi- cantly reduced tumor cell proliferation (Fig. 7G) and impaired colony-forming capacity (Fig. 7H), confirming that HOXC11 promotes the proliferative phenotype of ACC cells. Collectively, these findings highlight HOXC11 as a potential diagnostic biomarker and therapeutic target in ACC, particularly in patients with the high-risk MACCS1 subtype.
4 Discussion
ACC is a rare but highly aggressive malignancy that is often diagnosed at advanced or metastatic stages, which severely limits treatment effectiveness [37]. Its invasion and metastasis process involves the activation of multiple pro-migratory pathways by tumor- intrinsic gene mutations (such as TP53 and CTNNB1), as well as immunosuppression and angiogenesis induced by TME remodeling [38, 39]. While traditional therapies ben- efit early-stage patients, there remain no clearly defined molecular targets or effective drugs for advanced ACC, underscoring the urgent need for novel therapeutic strategies [40, 41].
In this study, we integrated transcriptomic, epigenetic, and genomic data to establish two robust ACC molecular subtypes: MACCS1, characterized by proliferative and poor- prognosis features, and MACCS2, enriched in metabolic and immune activity. This clas- sification was consistently validated across multiple independent cohorts and not only achieved statistical robustness but also reflected biologically coherent and clinically rel- evant distinctions. MACCS1 was dominated by cell cycle and DNA repair programs, while MACCS2 exhibited active immune engagement. Compared with prior ACC clas- sifications that relied on single-omics or transcriptomic clustering, our MACCS frame- work incorporates multiple molecular layers and provides clearer functional separation. Importantly, MACCS subtypes are linked to both prognosis and therapeutic response, enhancing their potential clinical utility.
In MACCS1, the five most frequently mutated genes were CTNNB1, TP53, TTN, MUC16, and CNTNAP5. By contrast, the mutational profile of MACCS2 was domi- nated by MUC16 and CMYA5. Although MUC16 or CMYA5 are not classic ACC driv- ers, their recurrent mutations in MACCS2 suggest functional contributions to tumor biology. MUC16 has been implicated in modulating cell adhesion, immune evasion, and chemoresistance in other cancers [42], while CMYA5 is involved in cytoskeletal organi- zation and could affect cell motility [43]. Their recurrent mutations in MACCS2 suggest possible contributions to TME interactions or metastatic potential, warranting further functional investigation.
Immune profiling revealed that MACCS2 had significantly higher expression of check- point molecules (CD274, CD247, PDCD1LG2) and increased infiltration of effector immune cells, forming an “immune-hot” phenotype with greater predicted sensitivity to ICIs. CD274 and PDCD1LG2, known ligands of PD-1, contribute to immune regula- tion by enhancing Treg function and preventing immune dysregulation [44-46]; while CD247 is critical for T cell receptor signaling and antigen recognition [47, 48]. Enrich- ment analysis confirmed that MACCS2 tumors were activated in CD8+ T cell effector and antigen presentation pathways, both essential for anti-tumor immunity [49-51]. By contrast, MACCS1 displayed high TMB but low immune activation, and was enriched in EMT and DNA repair pathways, suggesting that it escapes immune recognition through genomic stability maintenance and immunosuppressive mechanisms [52, 53]. The find- ings were significantly associated with a poor prognosis and shortened survival in this molecular subtype.
Drug sensitivity profiling identified subtype-specific therapeutic vulnerabilities. MACCS1 showed higher predicted sensitivity to axitinib, pazopanib, and several small molecules targeting oncogenic pathways such as PI3K/mTOR and RAS/RAF/MEK, con- sistent with its proliferative phenotype. Axitinib and pazopanib, which inhibit VEGFRs,
PDGFRs, FGFRs, and c-Kit, are already used clinically in renal cell carcinoma, lung can- cer, and other malignancies [54-56]. By contrast, MACCS2 was more sensitive to inhibi- tors of EGFR, MAPK, and JAK/STAT signaling, further highlighting its dependence on signal transduction pathways [57-59]. Although these in silico drug sensitivity predic- tions provide valuable hypotheses, they are based on cell line data and computational modeling. Differences between in vitro models and patient tumors, including microen- vironmental factors and drug metabolism, may limit direct clinical applicability. There- fore, experimental and clinical validation is essential before therapeutic application.
This study has several limitations. First, the analyses were based on publicly available datasets and a limited number of clinical samples, which may not fully capture the het- erogeneity of ACC. Second, the sample size in some omics layers was relatively small, potentially affecting statistical power. Third, although in vitro experiments were per- formed, in vivo validation in appropriate animal models or prospective clinical cohorts is still lacking. These factors should be addressed in future work to strengthen the trans- lational potential of the MACCS classification and its associated biomarkers.
This study also highlights HOXC11 as a candidate diagnostic and therapeutic target. Using random forest and Cox regression analysis, HOXC11 emerged as the most signifi- cant prognostic gene in MACCS1. High HOXC11 expression correlated with TP53 and PKHD1 mutations, increased chromosomal deletions (9p23, 9p21.3), and activation of cell cycle-related pathways such as the G2/M checkpoint and E2F targets, all consistent with a proliferative drive. Functionally, HOXC11 knockdown significantly reduced pro- liferation and colony formation in vitro, confirming its role in ACC tumor growth. Pre- vious studies in lung, breast, and colorectal cancers have shown that aberrant HOXC11 activity disrupts normal cell cycle regulation, DNA repair, and apoptosis, thereby pro- moting unchecked cell proliferation and tumor progression [60-62]. Our findings extend this evidence to ACC and support HOXC11 as a biomarker of poor prognosis and a potential therapeutic target, particularly in the high-risk MACCS1 subtype.
Despite its strengths, this study has limitations. First, the analyses relied on publicly available datasets with relatively small sample sizes, which may not fully capture the het- erogeneity of ACC. Second, some omics layers had limited statistical power. Third, while in vitro assays were performed, in vivo validation using animal models or clinical cohorts is still lacking. Addressing these limitations in future work will be crucial to strengthen the translational potential of the MACCS classification and its associated biomarkers.
4.1 Summarize
This study successfully constructed a molecular classification system for ACC through multi-omics consensus analysis, which profoundly revealed the significant differences in prognosis, biological behavior, genomic landscape and treatment sensitivity among different subtypes. Most importantly, this classification directly guides the stratification of treatment strategies: patients with the MACCS2 subtype may benefit from immune checkpoint inhibitor therapy, while patients with the MACCS1 subtype are more sensi- tive to specific targeted drugs. In addition, the identified key factor of high-risk subtype, HOXC11, is not only a powerful prognostic biomarker, but also a potential therapeu- tic target with preliminary functional verification. These findings not only deepen our understanding of the molecular mechanisms of ACC, a difficult-to-treat disease, but also provide a solid theoretical basis for the clinical implementation of precision
treatment based on molecular typing and the development of new targeted drugs, which is expected to significantly improve the clinical prognosis of ACC patients.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1007/s12672-025-03649-y.
Supplementary material 1.
Acknowledgements
Not applicable.
Author contributions
X.J. and X.W. jointly conceived and designed the study under the strategic guidance of Y.S. and G.L. X.J., X.W., and Z.W. developed the methodology, while X.J., Z.W., B.H., X.Z., and B.L. carried out the investigation. Data visualization was performed by X.W. and Z.W. Y.S. and G.L. supervised the entire project. X.J. and X.W. drafted the original manuscript, and Z.W., Y.S., and G.L. critically reviewed and revised it. All authors have read and approved the final version and accept responsibility for the integrity of the work.
Funding
No funding.
Data availability
Multi-omics data of ACC patients were obtained from The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.go v/), including transcriptome profiles, DNA methylation, somatic mutations, and corresponding survival information. Only patients with complete multi-omics and prognostic data were included in the analysis. For external validation, four independent ACC gene expression datasets were retrieved from the Gene Expression Omnibus (GEO) database: GSE10927 (n = 24), GSE33371 (n = 23), GSE70621 (n = 29), and GSE19750 (n = 22). As all data were derived from public repositories, no additional ethics approval was required. ☒
Declarations
Ethics approval and consent to participate
Not applicable.
Consent for publication
All authors have reviewed the manuscript and consented to its publication.
Competing interests
The authors declare no competing interests.
Received: 20 June 2025 / Accepted: 13 September 2025
Published online: 02 October 2025
References
1. Ghosh C, Hu J. Advances in translational research of the rare cancer type adrenocortical carcinoma. Nat Rev Cancer. 2023;23(12):805-24.
2. Ilanchezhian M, Varghese DG, Glod JW, et al. Pediatric adrenocortical carcinoma. Front Endocrinol (Lausanne). 2022;13:961650.
3. LIBé R, Huillard O. Adrenocortical carcinoma: diagnosis, prognostic classification and treatment of localized and advanced disease. Cancer Treat Res Commun. 2023;37:100759.
4. Sarrafan-Chaharsoughi Z, Yazdian Anari P, Malayeri AA, et al. Update on adrenocortical carcinoma. Urol Clin North Am. 2025;52(2):275-86.
5. Puglisi S, Calabrese A, Basile V, et al. New perspectives for mitotane treatment of adrenocortical carcinoma. Best Pract Res Clin Endocrinol Metab. 2020;34(3):101415.
6. Sun J, Huai J, Zhang W, et al. Therapeutic strategies for adrenocortical carcinoma: integrating genomic insights, molecular targeting, and immunotherapy. Front Immunol. 2025;16:1545012.
7. Kuhlen M, Schmutz M, Kunstreich M, et al. Targeting pediatric adrenocortical carcinoma: molecular insights and emerging therapeutic strategies. Cancer Treat Rev. 2025;136:102942.
8. Kimpel O, Dischinger U. Current evidence on local therapies in advanced adrenocortical carcinoma. Horm Metab Res. 2024;56(1):91-8.
9. Sinclair TJ, Alobuia WM, Gillis A, et al. Surgery for adrenocortical carcinoma: when and how? Best Pract Res Clin Endocrinol Metab. 2020;34(3):101408.
10. Araujo-Castro M, Pascual-Corrales E, Molina-Cerrillo J, et al. Immunotherapy in adrenocortical carcinoma: predictors of response, efficacy, safety, and mechanisms of resistance. Biomedicines. 2021;9(3):304.
11. Georgantzoglou N, Kokkali S, Tsourouflis G, et al. Tumor microenvironment in adrenocortical carcinoma: barrier to immu- notherapy success? Cancers (Basel). 2021. https://doi.org/10.3390/cancers13081798.
12. Wang SJ, Dougan SK, Dougan M. Immune mechanisms of toxicity from checkpoint inhibitors. Trends Cancer. 2023;9(7):543-53.
13. Zhang C, Zhang C, Wang H. Immune-checkpoint inhibitor resistance in cancer treatment: current progress and future directions. Cancer Lett. 2023;562:216182.
14. Ronchi CL, Altieri B, Kroiss M, et al. Next-generation therapies for adrenocortical carcinoma. Best Pract Res Clin Endocrinol Metab. 2020;34(3): 101434.
15. Hescheler DA, Hartmann MJM, Riemann B et al. Targeted therapy for adrenocortical carcinoma: A Genomic-Based search for available and emerging options. Cancers (Basel), 2022, 14(11).
16. Berruti A, Ferrero A, Sperone P, et al. Emerging drugs for adrenocortical carcinoma. Expert Opin Emerg Drugs. 2008;13(3):497-509.
17. Sukrithan V, Husain M. Emerging drugs for the treatment of adrenocortical carcinoma. Expert Opin Emerg Drugs. 2021;26(2):165-78.
18. Cremaschi V, Abate A, Cosentini D, et al. Advances in adrenocortical carcinoma pharmacotherapy: what is the current state of the art? Expert Opin Pharmacother. 2022;23(12):1413-24.
19. Ng L, Libertino JM. Adrenocortical carcinoma: diagnosis, evaluation and treatment. J Urol. 2003;169(1):5-11.
20. Shebrain S. Prediction of survival in adrenocortical carcinoma. J Invest Surg. 2022;35(5):1161-2.
21. VIëTOR CL, Schurink IJ, GRÜNHAGEN DJ, et al. Primary tumour resection in metastasised adrenocortical carcinoma. Endocr Relat Cancer. 2025. https://doi.org/10.1530/ERC-24-0056.
22. Lu Q, Nie R, Luo J, et al. Identifying immune-specific subtypes of adrenocortical carcinoma based on immunogenomic profiling. Biomolecules. 2023;13(1):104.
23. Cantini L, Zakeri P, Hernandez C, et al. Benchmarking joint multi-omics dimensionality reduction approaches for the study of cancer. Nat Commun. 2021;12(1):124.
24. Zheng S, Cherniack AD, Dewal N, et al. Comprehensive pan-genomic characterization of adrenocortical carcinoma. Can- cer Cell. 2016;29(5):723-36.
25. Lu X, Meng J, Zhou Y, Jiang L, Yan F. MOVICS: an R package for multi-omics integration and visualization in cancer subtyp- ing. Affiliations State Key Laboratory of Natural Medicines, Research Center of Biostatistics and Computational Pharmacy, China Pharmaceutical University, Nanjing 210009, China Department of Urology, The First Affiliated Hos, 2020, 36(22-23): 5539-41.
26. Chalise P, Fridley BL. Integrative clustering of multi-level’omic data based on non-negative matrix factorization algorithm. PLOS ONE. 2017;12(No.5):e0176278.
27. Tibshirani R, Walther G, Hastie T. Estimating the number of clusters in a data set via the gap statistic. Stanford University, USA, 2001, 63(2): 411- 23.
28. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2(Article). Harvard School of Public Health, Department of Biostatistics and Computational Biology, Dana Farber Cancer Institute and Department of Biostatistics, 450 Brookline Avenue, Boston, MA, United States; European Molecular Biology Laboratory, G, 2014, 15(12): 550.
29. Rooney MS, Shukla SA, Wu CJ, et al. Molecular and genetic properties of tumors associated with local immune cytolytic activity. Cell. 2015;160(1-2):48-61.
30. Xu Q, Xu H, Deng R, Wang Z, Li N, Qi Z, Zhao J, Huang W. Multi-omics analysis reveals prognostic value of tumor mutation burden in hepatocellular carcinoma. Affiliations The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University, No 109 Xueyuan West Road, Wenzhou, 325000, Zhejiang, China Zhejiang University School of Medi- cine, Hangzhou, 310009, Zh, 2021, Vol.21 (No.1): 342.
31. Xu F, Guan Y, Zhang P, et al. Tumor mutational burden presents limiting effects on predicting the efficacy of immune checkpoint inhibitors and prognostic assessment in adrenocortical carcinoma. Department Med Xi’an Jiaotong Univ Department Med Xi’an Jiaotong Univ Department Urol Second Affiliated Hosp Xi’an Jiaotong Univ Department Urol Second Affiliated H. 2022;221:1-14.
32. Li C, Mao Y, Liu Y, Hu J, Su C, Tan H, Hou X, Ou M. Machine learning-based integration develops a multiple programmed cell death signature for predicting the clinical outcome and drug sensitivity in colorectal cancer. Affiliations central laboratory. Second Affiliated Hosp Guilin Med Universi. 2025;36(1):1-18. Key Laboratory of Glucose and Lipid Metabolism DisordersThe Second Affiliated Hospital of Guilin Medical University Guangxi Health Commission.
33. Lim J, Jung h D, Lee K-M et al. Genome-scale metabolic modeling reveals a metabolic switch that restores sensitivity to anticancer chemotherapy in drug-resistant breast cancer cells. Cancer Res, 2023, 83(7).
34. LAMB J, CRAWFORD E D PECKD, et al. The connectivity map: using gene-expression signatures to connect small mol- ecules, genes, and disease. Science. 2006;313(5795):1929-35.
35. Brandau M, Kirsch V, Chernyakov D, et al. KO of ELAVL1 effects steroid synthesis in ACC cell line NCI-H295R. Univ Clin Halle Dept Intern Med 4 Halle Germany;Univ Witten Herdecke Inst Physiol Pathophysiology;Univ Clin Wurzburg Dept Intern Med 1 Wurzburg Ger. 2022;236:454.
36. Nishi H, Arai H. NCI-H295R, a human adrenal cortex-derived cell line, expresses purinergic receptors linked to Ca2+- mobilization/influx and cortisol secretion. PLoS ONE. 2013;8(8):e71022.
37. Gara SK, Lack J, Zhang L, et al. Metastatic adrenocortical carcinoma displays higher mutation rate and tumor heterogene- ity than primary tumors. Nat Commun. 2018;9(1):4172.
38. Fassnacht M, LIBé R, Kroiss M, et al. Adrenocortical carcinoma: a clinician’s update. Nat Rev Endocrinol. 2011;7(6):323-35.
39. Sasano H, Suzuki T, Moriya T. Recent advances in histopathology and immunohistochemistry of adrenocortical carcinoma. Endocr Pathol. 2006;17(4):345-54.
40. Ali AE, Raphael SJ. Functional oncocytic adrenocortical carcinoma. Endocr Pathol. 2007;18(3):187-9.
41. Bertherat J, Coste J, Bertagna X. Adjuvant mitotane in adrenocortical carcinoma. N Engl J Med. 2007;357(12):1256-7.
42. Chen X, Sandrine IK, Yang M, et al. MUC1 and MUC16: critical for immune modulation in cancer therapeutics. Frontiers in immunology. 2024;15:1356913.
43. Lu F, Ma Q, Xie W, et al. CMYA5 establishes cardiac dyad architecture and positioning. Nat Commun. 2022;13(1):2185.
44. Alexander PG, Mcmillan DC, Park JH. A meta-analysis of CD274 (PD-L1) assessment and prognosis in colorectal cancer and its role in predicting response to anti-PD-1 therapy. Crit Rev Oncol Hematol. 2021;157:103147.
45. Dermani FK, Samadi P, Rahmani G, et al. PD-1/PD-L1 immune checkpoint: potential target for cancer therapy. J Cell Physiol. 2019;234(2):1313-25.
46. Chen Q, Liu M, Zhao P, et al. Prognostic significance of PDCD1LG2 expression in pan-cancer and its relationship with the immune microenvironment. Asian J Surg. 2024;47(12):5288-90.
47. Aguinaga-Barrilero A, Castro-Sánchez P, Juárez I, et al. Defects at the Posttranscriptional Level Account for the Low TCRZ Chain Expression Detected in Gastric Cancer Independently of Caspase-3 Activity. J Immunol Res. 2020;2020:1039458.
48. Ferrandina G, Ranelletti FO, Legge F, et al. Celecoxib up-regulates the expression of the zeta chain of T cell receptor com- plex in tumor-infiltrating lymphocytes in human cervical cancer. Clin Cancer Res. 2006;12(7 Pt 1):2055-60.
49. Huang L, Wang D, Xu M, et al. Mixed radiation with different doses induces CCL17 to recruit CD8(+)T cell to exert anti- tumor effects in non-small cell lung cancer. Front Immunol. 2024;15:1508007.
50. Wang Q, Qin Y, Li B. CD8(+) T cell exhaustion and cancer immunotherapy. Cancer Lett. 2023;559:216043.
51. Yang K, Halima A, Chan TA. Antigen presentation in cancer - mechanisms and clinical implications for immunotherapy. Nat Rev Clin Oncol. 2023;20(9):604-23.
52. Nachiyappan A, Gupta N. EHMT1/EHMT2 in EMT, cancer stemness and drug resistance: emerging evidence and mecha- nisms. FEBS J. 2022;289(5):1329-51.
53. Chalmers ZR, Connelly CF. Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden. Genome Med. 2017;9(1):34.
54. Ferrarotto R, Sousa LG, Feng L, et al. Phase II clinical trial of axitinib and avelumab in patients with recurrent/metastatic adenoid cystic carcinoma. J Clin Oncol. 2023;41(15):2843-51.
55. Plimack ER, Powles T. Pembrolizumab plus axitinib versus sunitinib as first-line treatment of advanced renal cell carcinoma: 43-month follow-up of the phase 3 KEYNOTE-426 study. Eur Urol. 2023;84(5):449-54.
56. Rini BI, Plimack ER, Stus V, et al. Pembrolizumab plus axitinib versus Sunitinib for advanced Renal-Cell carcinoma. N Engl J Med. 2019;380(12):1116-27.
57. Nussinov R, Yavuz BR, Jang H. Anticancer drugs: how to select small molecule combinations? Trends Pharmacol Sci. 2024;45(6):503-19.
58. Wu Q, Qian W, Sun X, et al. Small-molecule inhibitors, immune checkpoint inhibitors, and more: FDA-approved novel therapeutic drugs for solid tumors from 1991 to 2021. J Hematol Oncol. 2022;15(1):143.
59. Wang P, Jin X. Recent advances in small molecule prodrugs for cancer therapy. Anticancer Agents Med Chem. 2014;14(3):418-39.
60. Han L, Gong F, Wu X, et al. Comprehensive characterization of PKHD1 mutation in human colon cancer. Cancer Med. 2024;13(1):e6796.
61. Shang T, Chen X, Xue H, et al. The PKHD1 gene inhibits tumor proliferation and invasion in intrahepatic cholangiocarci- noma by activating the Notch pathway. Int J Med Sci. 2024;21(14):2655-63.
62. Giacomelli AO, Yang X, Lintner RE, et al. Mutational processes shape the landscape of TP53 mutations in human cancer. Nat Genet. 2018;50(10):1381-7.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.