Cohort collection and multi-omics profiling of pediatric adrenocortical tumors
To investigate the diverse molecular landscape of pACTs, we enrolled 109 children and adolescents from the German Malignant Endocrine Tumors (MET) Registry diagnosed with a pACT between 1997 and 2024. This cohort included cases of pACCs (n = 60), pACAs (n = 35) and pUMPs (n = 14) according to the Wieneke criteria. Given the geographic variability with respect to tumor predisposition, we integrated previously published cohorts from Clay et al. (n = 60) and Bueno et al. (n = 57), thus including cohorts enriched for the Brazilian founder variant. Overall, this resulted in a cohort of 214 pACT and 12 healthy adrenal cortex control samples. The mean age at diagnosis across the entire cohort was 5.3 years with a female to male ratio of 2.3:1. 75.2% of patients presented with virilization (n = 89), Cushing’s syndrome (n = 20), or both (n = 52).
Comprehensive DNA methylation profiling across the entire cohort helped define molecular subgroups and improve risk stratification in pACTs, establishing the foundation for a biologically meaningful classification. Aiming not only for an improved stratification system but also to unravel the epitranscriptomic characteristics, we enriched our dataset by single nucleus RNA-sequencing (n = 29) and snATACseq (n = 7) (Fig. 1A).
Fig. 1: DNA methylation–based stratification of pediatric adrenocortical tumors identifies four clinically distinct subgroups.
A Overview of the project presented in this paper. Created in BioRender. Fincke, V. (2026) https://BioRender.com/5ujxlr5. B Unsupervised clustering of DNA methylation data using Consnsus Clustering identifies four molecular subgroups (Low Risk 1, Low Risk 2, Low Risk 3, and High Risk) in 214 pediatric ACTs. Red indicated samples always cluster together; blue color indicates samples never cluster together (total n = 214; HR, n = 36; LR1, n = 104; LR2, n = 34; LR3, n = 40). C UMAP projection of DNA methylation profiles colored by molecular subgroup (total n = 214; HR, n = 36; LR1, n = 104; LR2, n = 34; LR3, n = 40). D Kaplan–Meier analysis of overall survival across histopathological diagnoses (pACA, n = 82; pACC, n = 92; pUMP, n = 40; log-rank test). E Kaplan–Meier analysis of overall survival across methylation-based subgroups (HR, n = 36; LR1, n = 104; LR2, n = 34; LR3, n = 40; log-rank test). F COG/IPACTR stage at diagnosis across methylation-based subgroups (stage 1, n = 75; stage 2, n = 67; stage 3, n = 27; stage 4, n = 36; stage not available, n = 9; chi-square test). G Age at diagnosis (years) by molecular subgroup (HR, n = 35; LR1, n = 102; LR2, n = 34; LR3, n = 40; age at diagnosis was available for 211 patients). Violin plots show the kernel density distribution of the data; embedded box plots indicate the median (center line), the 25th and 75th percentiles (box bounds), and whiskers extending to the most extreme value within 1.5 × the interquartile range of the box, which define the plotted minima and maxima. Groups were compared by two-sided Kruskal–Wallis test (P = X), followed by two-sided pairwise Dunn’s post hoc tests with Benjamini–Hochberg correction for multiple comparisons. Adjusted P values: LR1 vs LR2, Padj = 0.129; LR1 vs LR3, Padj = 0.018; LR1 vs HR, Padj = X; LR2 vs LR3, Padj = 0.0019; LR2 vs HR, Padj = X; LR3 vs HR, Padj = 1.64 × 10⁻⁴. Source data are provided with this paper.
DNA methylation profiling reveals four distinct subgroups with prognostic significance
We performed consensus clustering based on DNA methylation profiles, which identified four distinct subgroups (Fig. 1B). The robustness of this clustering was supported by clustering metrics, including the relative change in the area under the cumulative distribution function (CDF) curve at k = 4 (Supplementary Fig. S1A), confirming the stability of the four-group solution. Reduction of the high-dimensional methylation data using Uniform Manifold Approximation and Projection (UMAP) demonstrated separation of the identified subgroups (Fig. 1C). Subgroup identity was initially derived from unsupervised consensus clustering of DNA methylation profiles; risk-based designations were subsequently assigned based on the correlation of each subgroup with clinical outcome data available for our cohort. These subgroups were designated High Risk (HR), Low Risk 1 (LR1), Low Risk 2 (LR2), and Low Risk 3 (LR3), reflecting their association with patient overall survival (OS). The five-year OS for patients in the HR subgroup was 25%, which is significantly lower than for those in the Low Risk subgroups (p < 0.05; log-rank test; (HR: 25% [95% CI: 13.7%–46.5%]; LR1: 91% [95% CI: 84.7%–97.7%]; LR2: 96% [95% CI: 89.8%–100%]; LR3: 97% [95% CI: 90.5–100%])) (Fig. 1D).
Benchmarking these subgroups against the commonly used histopathological classification into pACC, pACA, and pUMP has prognostic relevance (p < 0.05; log-rank test (Fig. 1E)), our DNA methylation-based subgroups more accurately identified patients at high risk, indicating superior prognostic resolution.
Among the four subgroups, the largest proportion of samples was assigned to LR1 (104/214, 50.9%), followed by LR3 (40/214, 18.7%), HR (36/214, 16.8%), and LR2 (34/214, 13.6%) (Supplementary Fig. S1B). This refined classification based on DNA methylation profiles allows for improved risk stratification, particularly for pACC and pUMP patients the latter of which are diagnostically challenging. Methylation-based subgrouping showed prognostic significance in pACC, whereas no significant association with outcome was observed in the UMP and ACA diagnostic groups (Supplementary Fig. S2).
Comparison with previously established DNA methylation-based subgroup classifications by Clay et al. (A1 and A217,) and Bueno et al. (pACT-1 and pACT-218,) revealed partial overlap with our clustering (Supplementary Fig. S1B): To benchmark our DNA methylation–based classification against these prior subgrouping efforts, we mapped their previously defined cohorts (Clay A1/A2 and Bueno pACT-1/2) onto our risk groups. Notably, a substantial proportion of tumors previously classified as HR (Clay A1 and Bueno pACT-1) corresponded to our HR subgroup: 8 of the 10 tumors classified as pACT-1 by Bueno et al. were assigned to our HR group, with 2 classified as LR2. Similarly, of the 15 Clay A1 tumors available, 7 mapped to our HR group, while the remainder were split between LR2 (n = 6) and LR3 (n = 2) (Supplementary Data S1).
These findings demonstrate a limited intersection between previously proposed HR DNA methylation classes and our HR subgroup – particularly for the pACT-1 classification by Bueno et al. At the same time, the distribution of Clay A1 tumors across multiple subgroups in our dataset underscores the added biological granularity and refinement achieved by our four-group model: From the 8 samples that were identified as LR from the A1 group by DNA methylation, 87.5% (7 out of 8) survived the disease.
Patients in the HR subgroup presented with a more advanced tumor stage at diagnosis (Fig. 1F) and were significantly older compared to patients in the three LR subgroups (Fig. 1G). Notably, no pACA case was classified into the HR subgroup, whereas three pUMP tumors were assigned to this subgroup (Supplementary Fig. S1C). Sex distribution differed significantly across DNA methylation subgroups (χ²3 = 11.53, p < 0.05). While in the HR group sex was equally distributed, the different LR groups showed uneven female-to-male ratios (Supplementary Fig. S1D).
Pediatric adrenocortical tumors frequently present with signs of hormonal excess reflecting the steroidogenic activity of the tumor. Virilization, resulting from androgen overproduction, is the most common presentation and manifests as premature pubic hair, clitoral or penile enlargement, and accelerated growth. Cushing syndrome arises from excess cortisol secretion and is characterized by weight gain, facial rounding, hypertension, and growth retardation, while hypertension-related presentations are associated with mineralocorticoid or mixed steroid excess. The endocrine phenotype of the patients at presentation varied significantly across methylation subgroups (Fisher’s exact test with Monte Carlo simulation, p < 0.05; 100,000 replicates, Supplementary Fig. S1E). Virilization was more common in LR1, whereas Cushing syndrome was enriched in HR and rare in LR1. LR3 was associated with hypertension-related presentations. These findings support that HR tumors are more likely to present with Cushing syndrome or without endocrinological symptoms, and rarely with virilization.
The Ki-67 proliferation index is a widely used marker of tumor cell proliferation, assessed here by immunohistochemistry on tumor tissue sections, while MKI67 transcript levels were derived from bulk RNA sequencing data. Neither measure differed significantly across DNA methylation-defined subgroups (Supplementary Fig. S1F, G).
A high Ki-67 index (>20%) correlated with inferior overall survival (log-rank p < 0.05), but its prognostic performance remained modest, with time-dependent AUCs decreasing from 0.76 at 1 year to 0.65 at 10 years (Supplementary Fig. S3). In contrast, DNA methylation-based classification demonstrated superior and steadily improving prognostic accuracy, with AUCs increasing from 0.79 at 1 year after diagnosis to 0.92 at 10 years after diagnosis. While pS-GRAS retained robust prognostic capacity (Supplementary Fig. S4), it demonstrated slightly reduced long-term discriminatory power compared to DNA methylation classification, with lower AUCs at 10 years after diagnosis. While the COG-IPACTR classification, although highly predictive in the short term (AUC 0.87 at 1 year), showed a moderate decline at later time points (AUC 0.89 at 10 years). The Wieneke classification consistently underperformed relative to these models, with AUCs peaking at ~0.80 and declining to 0.72 by 10 years after diagnosis.
Importantly, the advantage of DNA methylation based subgrouping was particularly evident in clinically challenging cases classified as COG/IPACTR stage 2 or 3, i.e., larger tumors with or without local invasion whose clinical course is hard to predict, as they are non-metastatic upfront but have a tendency to recur (Supplementary Fig. 5). In this subgroup, DNA methylation achieved markedly higher discriminative power compared to both ps-GRAS and COG-IPACTR classifications, 0.93 compared to 0.86 and 0.66 at 10 years after diagnosis (Supplementary Fig. S3).
To provide molecular subgroup assignment in individual cases without requiring cohort-wide methylation clustering, we developed a random forest–based classifier trained on DNA methylation data. This tool is designed to facilitate the application of our subgroup framework in both research and clinical contexts, including settings where comprehensive reference cohorts may not be available. The classifier demonstrated high accuracy (0.934) in predicting subgroup membership, as confirmed by three nested cross-validation, yielding a high area under the receiver operating characteristic (ROC) curve (AUC) of 0.989 and a low misclassification rate of 0.066 (Supplementary Fig. S6). In the future, this online classifier can support clinical decision-making by enabling objective, molecularly informed risk stratification in diagnostic workflows and multicenter clinical trials.
High-risk pediatric adrenocortical tumors exhibit elevated CpG island methylation and increased copy number variability
Analysis of methylation levels of the subgroups in comparison to 12 healthy adrenal cortex samples17 and 79 adult ACTs16 revealed CpG island hypermethylation in HR pACTs with a median methylation level of 0.62 compared to 0.2 in LR1 (Fig. 2A). Methylation levels differ significantly among the HR subgroup and all LR subgroups (Dunn’s test, p < 0.05), but not when comparing the pACT HR to the adult CIMP-high subgroup (median methylation level = 0.75). In UMAP projection the HR subgroup exhibits notable similarity to the adult CIMP-High subgroup (Supplementary Fig. S7A, B).
Fig. 2: High-risk pACTs exhibit enhanced CpG island methylation, genomic instability, and enriched TP53 mutations.
A CpG island methylation levels (top 1% most variable probes) across pACT methylation subgroups (LR1, n = 104; LR2, n = 34; LR3, n = 40; HR, n = 36), adult adrenocortical carcinoma (ACC) CpG island methylator phenotype (CIMP) categories from the TCGA cohort (CIMP-high, n = 20; CIMP-intermediate, n = 27; CIMP-low, n = 32) and healthy adrenal cortex controls (n = 12); total n = 305. Violin plots show the kernel density distribution of the data; embedded box plots indicate the median (center line), the 25th and 75th percentiles (box bounds) and whiskers extending to the most extreme value within 1.5 × the interquartile range of the box, which define the plotted minima and maxima. Groups were compared by two-sided Kruskal–Wallis test, followed by two-sided Dunn’s post hoc tests with Benjamini–Hochberg correction across all 28 pairwise comparisons. Source data are provided with this paper. Adjusted p values: LR1 vs. LR2, p.adj = 1.5 × 10⁻⁸; LR1 vs. LR3, p.adj = 4.9 × 10⁻⁶; LR1 vs. HR, p.adj = 6.7 × 10⁻²⁵; LR1 vs. CIMP-high, p.adj = 5.0 × 10⁻²³; LR1 vs. CIMP-intermediate, p.adj = 2.8 × 10⁻¹⁸; LR1 vs. CIMP-low, p.adj = 1.2 × 10⁻⁴; LR1 vs. Healthy, p.adj = 0.263; LR2 vs. LR3, p.adj = 0.227; LR2 vs. HR, p.adj = 4.5 × 10⁻⁴; LR2 vs. CIMP-high, p.adj = 1.0 × 10⁻⁵; LR2 vs. CIMP-intermediate, p.adj = 0.0053; LR2 vs. CIMP-low, p.adj = 0.158; LR2 vs. Healthy, p.adj = 0.0201; LR3 vs. HR, p.adj = 1.0 × 10⁻⁶; LR3 vs. CIMP-high, p.adj = 2.5 × 10⁻⁸; LR3 vs. CIMP-intermediate, p.adj = 4.5 × 10⁻⁵; LR3 vs. CIMP-low, p.adj = 0.734; LR3 vs. Healthy, p.adj = 0.142; HR vs. CIMP-high, p.adj = 0.158; HR vs. CIMP-intermediate, p.adj = 0.672; HR vs. CIMP-low, p.adj = 9.8 × 10⁻⁷; HR vs. Healthy, p.adj = 1.0 × 10⁻⁶; CIMP-high vs. CIMP-intermediate, p.adj = 0.0949; CIMP-high vs. CIMP-low, p.adj = 2.5 × 10⁻⁸; CIMP-high vs. Healthy, p.adj = 2.5 × 10⁻⁸; CIMP-intermediate vs. CIMP-low, p.adj = 3.0 × 10⁻⁵; CIMP-intermediate vs. Healthy, p.adj = 1.1 × 10⁻⁵; CIMP-low vs. Healthy, p.adj = 0.214. *p.adj <0.05; **p.adj <0.01; ***p.adj <0.001; ****p.adj <0.0001; ns, not significant. Source data are provided as a Source Data file. B Copy number alteration frequency across chromosomes stratified by methylation subgroup. Gains and losses are shown as the percentage of samples affected per chromosome (total n = 214, HR, n = 36; LR1, n = 104; LR2, n = 34; LR3, n = 40). C Frequency of germline TP53 mutations across methylation-based subgroups (HR, n = 26; LR1, n = 76; LR2, n = 15; LR3, n = 24)17,18. Pairwise two-sided Fisher’s exact tests with Benjamini-Hochberg correction for multiple comparisons: LR1 vs. LR2, p.adj = 0.00176; LR2 vs. LR3, p.adj = 0.00996; LR2 vs. HR, p.adj = 0.0462; LR1 vs. LR3, p.adj = 0.809 (ns); LR1 vs. HR, p.adj = 0.252 (ns); LR3 vs. HR, p.adj = 0.488 (ns). * p.adj <0.05, ** p.adj <0.005. Percentages indicate subgroup-specific germline TP53 mutation rates. Source data are provided as a Source Data file. D Distribution and types of somatic TP53 mutations across functional domains of the TP53 gene. E Frequency of somatic mutations in known ACT-associated genes (e.g., TP53, CTNNB1, IGF2, ATRX) among patients with available whole-exome sequencing (n = 45: HR, n = 11; LR1, n = 22; LR2, n = 8; LR3, n = 4). F Bar plot displaying the frequency of somatic CTNNB1 mutations per subgroup (n = 159: HR, n = 30; LR1, n = 83; LR2, n = 21; LR3, n = 25); two-sided pairwise Fisher’s exact tests, Benjamini-Hochberg corrected. Adjusted p-values: HR vs. LR1, p.adj = 0.0754 (ns); HR vs. LR2, p.adj = 0.00135 (**); HR vs. LR3, p.adj = 1.000 (ns); LR1 vs. LR2, p.adj <0.0001 (****); LR1 vs. LR3, p.adj = 0.0863 (ns); LR2 vs. LR3, p.adj = 0.00179 (**). ns, not significant (**p.adj <0.01, ****p.adj <0.0001). Source data are provided as a Source Data file.
Methylation-derived copy number analysis revealed distinct subgroup-specific patterns of chromosomal instability (Fig. 2B). HR tumors exhibited a high burden of recurrent chromosomal alterations, as determined by GISTIC analysis. Specifically, we identified six significant copy number gains and twenty losses. Amplified regions included oncogenic loci such as MYC, CDK4, MDM2, and AURKC, while recurrent deletions encompassed key tumor suppressor genes including BRCA2, CHEK2 and TP73. However, expression levels of these genes showed highly heterogeneous patterns across HR patients and did not consistently correlate with the underlying copy number status, pointing to high inter-patient variability and suggesting that additional regulatory mechanisms beyond copy number may govern their expression in this subgroup (Supplementary Fig. S8A). In contrast, LR2 tumors showed remarkable genomic stability, with only a single recurrent deletion on chromosome 4q35.2 (Supplementary Fig. S8B-E).
Subgroup-specific differences in genetic alterations were evident: LR2 tumors exhibited a significantly lower frequency of germline TP53 variants (13.3%) than each of the other three subgroups, whereas frequencies in HR, LR1, and LR3 ranged from 50.0% to 65.8% and did not differ significantly from one another (pairwise two-sided Fisher’s exact tests with Benjamini-Hochberg correction for multiple comparisons; LR2 vs. LR1, p.adj = 0.00175; LR2 vs. LR3, p.adj = 0.00995; LR2 vs. HR, p.adj = 0.0462; all other pairwise comparisons ns; Fig. 2C). At the somatic level, TP53 mutations were more evenly distributed across subgroups and were also present in LR2 (13.3%), though at lower frequency than in HR (23.1%). These somatic alterations spanned key functional domains of TP53 – including the DNA-binding and tetramerization regions – and included missense, frameshift, and splice-site variants (Fig. 2D).
Further somatic mutational analysis revealed TP53 and CTNNB1 (28%, respectively) as the most frequently altered genes, with infrequent mutations in ATRX, DAXX, TERT, and IGF2 (Fig. 2E, Supplementary Fig. S9, Supplementary Data S2). Striking differences in the distribution of somatic CTNNB1 mutations were observed across subgroups, with LR2 exhibiting a significantly elevated mutation frequency (76.2%) compared to all other subgroups (5.1–24%; p < 0.5) (Fig. 2F). Expression of genes known to be mutated in adult ACCs, such as ATRX, DAXX or TERT, do not show any differences in RNA expression between subgroups (Supplementary Fig. S8).
The composition of pediatric adrenocortical tumors mimic the physiological, zonal differentiation of the physiological adrenal cortex
As the genomic differences do not account for all the biological differences among the subgroups, we hypothesized that cellular composition among the subgroups differs substantially. We therefore performed single-nucleus RNA sequencing (snRNA-Seq) on 18 pACT samples representing all DNA methylation-based risk groups including 11 pACCs, 6 pACAs, and 1 pUMP (Fig. 3A, B). Notably, samples of the same subgroup clustered together at this data level. Following quality control measures, transcriptional profiles from 75,515 single cells, with a median of 1,297 genes per cell, were retained for subsequent analysis. Using established markers for cells of the tumor microenvironment (TME)—including CD68 and CD163 for macrophages, ACTA2 and PDGFRB for fibroblasts, CD3D and CD79A for lymphoid cells, and PECAM1 for endothelial cells—distinct clusters of these populations were identified, alongside multiple tumor cell clusters (Fig. 3C, Supplementary Data S3).
Fig. 3: Single-Nucleus RNA sequencing reveals adrenal zonation and cell type diversity across pACT zubgroups.
A Clinical annotation of each sample, including risk group, gender, age category, clinical presentation, and survival status. B UMAP projection of snRNA-seq data from 18 pACT samples ((HR: 4 pACC; LR1: 4 pACC, 4 pACA; LR2: 1 pACC, 2 pACA, 1 pACx; LR3: 2 pACC)), colored by patient ID for n = 75,515 nuclei. C UMAP embedding of all cells (n = 75,515 cells, n = 18 samples) colored by inferred cell type identity, including glomerulosa-like, zF/zR-like Cluster I and II, macrophages, lymphoid cells, endothelial cells, fibroblasts, and cycling cells. D Expression of key adrenal steroidogenic genes across the UMAP space. Genes shown include DAB2, SYTL2, CYP11A1, CYP21A2, STAR, CYP17A1, CYP11B1, and HSD3B2. Diagram (center) summarizes the biosynthetic zonation of the adrenal cortex and corresponding gene functions. Created in BioRender. Fincke, V. (2026) https://BioRender.com/34flu07. E Pseudotime projection of tumor cells (n = 53,943 tumor cells, n = 18 samples), indicating a potential differentiation trajectory from glomerulosa-like cells toward zF/zR-like clusters. F Proportional representation of tumor cell clusters across molecular subgroups (n = 18 samples divided across n = 4 subgroups: HR n = 4, LR1 n = 8, LR2 n = 4, LR3 n = 2). Bubble size reflects the relative abundance of each cluster per subgroup.
To characterize the TME across risk groups, we systematically quantified the proportional abundance of immune and stromal cell populations in each sample (Supplementary Fig. S10). This analysis revealed significant differences in macrophage and lymphoid cell proportions across methylation-defined risk groups (Kruskal-Wallis, p = 0.047 and p = 0.049, respectively). HR tumors showed a marked depletion of macrophages compared to LR1 and LR2 tumors (median 1.5% vs. 8.1% and 10.3%, respectively), and were similarly depleted of lymphoid cells (median 0.2% in HR vs. 4.2% in LR1). Endothelial cell and fibroblast proportions did not differ significantly across subgroups (p = 0.329 and p = 0.265, respectively), although endothelial cells were numerically lower in HR tumors (median 3.1%) compared to LR1 (median 16.4%). Collectively, these findings suggest that HR pACTs are characterized by a markedly immunosilenced microenvironment, with reduced macrophage infiltration and lymphoid cell abundance, which may contribute to immune evasion and the more aggressive clinical behavior of this subgroup.
Analysis of the tumor subclusters revealed that their transcriptional profiles closely resembled those of the physiological adrenal cortex, as evidenced by the expression of zone-specific marker genes. Two clusters clearly displayed features resembling the zF and zR, distinguished by the expression of CYP17A1 and CYP11A1, respectively – key steroidogenic enzymes that serve as established markers for these adrenal cortical zones (Fig. 3D). Another tumor cluster exhibited high expression of genes typically upregulated in the outermost adrenal cortex zone, zG, including CYP11B2 and DAB2. This cluster was therefore classified as the zG-like cluster. Additionally, a distinct cluster of cycling tumor cells was identified, characterized by elevated S.Score and G2M.Score in cell cycle analysis (Supplementary Data S4). Thus, the transcriptome of pACTs parallels the physiological cell types present in the adrenocortex.
This observation is further corroborated by pseudotime analysis suggesting a differentiation trajectory, originating from zG-like cells and progressing toward the zF/zR-like clusters, thereby mimicking the differentiation pattern of the healthy adrenal cortex (Fig. 3E). To investigate how this transcriptional hierarchy varies across molecular subgroups, we quantified the representation of each transcriptional cell state within the DNA methylation-based subgroups (Fig. 3F). We found that the HR subgroup was characterized by a nearly balanced distribution of cells in the zG-, zF/zR-like states in contrast to the more skewed cell state distributions observed in the LR subgroups. These relative enrichment and co-existence of all three major adrenal cortical-like lineages in HR tumors suggest a broader differentiation spectrum and increased plasticity within this group and suggests that each of these cell populations may fulfill a distinct role in tumorigenesis.
Strikingly, HR tumors exhibited the highest cellular heterogeneity, with an average Shannon Index of 1.01, compared to values ranging from 0.61 to 0.87 in the other subgroups. This elevated diversity supports the notion that HR tumors are not only transcriptionally distinct but also compositionally more complex than the LR groups.
Because bulk DNA methylation profiling was performed on heterogeneous tumor samples, we next asked whether subgroup-specific methylation differences might reflect variation in cellular composition rather than tumor-intrinsic epigenetic states. To address this, we integrated bulk methylation data with matched snRNA-seq profiles. Global methylation levels were not associated with tumor cell fraction estimated from snRNA-seq (Spearman ρ = 0.06, p = 0.82), nor with tumor pseudobulk transcriptional activity (ρ = 0.02, p = 0.95). In addition, no significant correlations were observed between global methylation and the proportional abundance of individual tumor or microenvironmental clusters (Supplementary Fig. S11). These findings suggest that methylation differences between molecular subgroups are unlikely to be driven by compositional effects and instead reflect tumor-intrinsic epigenetic programs.
This prompted us to investigate the regulatory models which underlie the different subgroups by using NMF based decomposition.
Canonical WNT-signaling, emanated by the ZG-like population, represents a major constituent of the HR subgroup
NMF based decomposition enables the deconvolution of complex gene expression matrices into interpretable components, each representing a distinct transcriptional program active across a subset of cells, thereby revealing patterns that are not necessarily aligned with predefined cell clusters or known lineages. NMF was applied independently to each sample, which revealed seven shared transcriptional metasignatures after hierarchical clustering (Fig. 4A, Supplementary Data S5). These metasignatures (MS) were differentially distributed among samples across the DNA methylation-based risk groups (Fig. 4B). To characterize each metasignature, we performed Gene Ontology (GO) enrichment analysis and examined the top overexpressed genes. MS4 (Cell Cycle Activation) showed strong enrichment for mitotic processes (e.g., chromosome segregation, nuclear division) and was defined by genes such as NUF2, FAM166B, and LSS, indicating a highly proliferative cell state (Fig. 4C). In contrast, MS1 (Steroidogenesis) included canonical steroidogenic markers such as SCARB1, CYP11B1, and STAR. MS2 and MS3 captured distinct transcriptional stress programs: MS2 (Stress and Hypoxia) was enriched for corresponding genes (e.g., HIF3A, PLAGL1, FAM166B), consistent with adaptation to a nutrient-deprived or hypoxic microenvironment. In contrast, MS3 (Steroidogenic Stress Response) reflected a steroidogenic stress response, marked by upregulation of STAR, FKBP5, SULT2A1, and HSP90AA1, suggesting sustained hormonal output and compensatory mechanisms supporting tumor cell survival under chronic steroidogenic pressure. MS5 (Resistance and Differentiation-Stabilizing) included COL11A1 and RAB38, which have been implicated in extracellular matrix remodeling and therapy resistance mechanisms. Further, MS6 (Immune Modulation) was defined by genes such as ZNF117 and CHST15, while MS7 (ECM Structural Integrity) involved ERV3-1, GRIN2C, and ZNF431, collectively suggesting roles in tumor–immune interactions and stromal architecture.
Fig. 4: Transcriptional metasignatures reveal functional heterogeneity and differential WNT pathway engagement across pACT Subgroups.
A Heatmap of pairwise correlations among seven transcriptional metasignatures derived from non-cycling tumor cells using non-negative matrix factorization (NMF) across 18 pACT samples. B Bar plots showing sample-level proportions of metasignature expression. Each bar represents one tumor sample, colored by metasignature contribution. C GO term enrichment analysis for Metasignature 4 (n = 30 genes), annotated as “Proliferative and Cell Cycle Activation,” showing top enriched pathways and gene counts. Enrichment was tested against the default genome-wide background of clusterProfiler using a one-sided hypergeometric test (clusterProfiler enrichGO), with p values adjusted for multiple comparisons by the Benjamini-Hochberg method. D Enrichment heatmap showing the prevalence of metasignatures in the tumor cell clusters in the molecular risk groups(HR: 4 pACC; LR1: 4 pACC, 4 pACA; LR2: 1 pACC, 2 pACA, 1 pACx; LR3: 2 pACC). E Dot plot showing ligand–receptor interactions inferred by LIANA across tumor clusters (zF/zR-like Cluster I n = 17,970 cells, zF/zR-like Cluster II n = 22,619, zG-like Cluster n = 12,354 cells. Zona fasciculata (zF), zona reticularis (zR), zona glomerulosa (zG). F Immunohistochemical staining for CTNNB1 in representative tumor samples from each risk group (HR, n = 7; LR1, n = 4; LR2, n = 3; LR3, n = 3 tumors stained). High Risk and Low Risk 2 tumors show strong nuclear localization, while Low Risk 1 and Low Risk 3 exhibit primarily membranous staining, indicating subgroup-specific activation of canonical WNT signaling. The images shown are representative of the staining pattern observed in all stained samples of the respective subgroup. Scale bars indicate 50 µm.
Importantly, the distribution of these metasignatures also reflected the clinical presentation associated with each methylation subgroup (Supplementary Fig. S1E). For instance, LR3 tumors—enriched for MS1 (Steroidogenesis) and MS3 (Steroidogenic Stress Response)—showed a significant overrepresentation of virilization symptoms, consistent with enhanced or dysregulated androgen biosynthesis. Conversely, HR tumors, characterized by MS4 (Cell Cycle Activation) and MS6 (Immune Modulation), were associated with Cushing’s syndrome, aligning with the proliferative and cortisol-producing profile of these tumors. Finally, LR3 tumors—clinically marked by frequent virilization—were enriched for MS1 (Steroidogenesis) and MS5 Resistance and Differentiation Stabilizing, suggesting a more differentiated and hormonally specialized transcriptional state.
The metasignatures exhibited differential distribution not only within risk groups but also between tumor cell populations. Notably, the MS4 (Cell Cycle Activation) was strongly—and almost exclusively—present in the zG-like cluster within the HR subgroup (Fig. 4D). While LR2 and LR3 tumors were predominantly defined by a limited number of transcriptional programs, LR1 and HR tumors exhibited broader metasignature diversity. In the HR subgroup—which exhibited a balanced distribution of all major transcriptional clusters—the concurrent presence of diverse metasignatures suggests that inter-cluster communication may play a critical role in driving tumor progression.
To dissect the intercellular communication dynamics of zG-like tumor cells in different risk subgroups, we performed ligand-receptor interaction analysis stratified by subgroups. In HR tumors, zG-like cells exhibited a prominent WNT signaling axis, characterized by extensive ligand-receptor interactions including WNT1, WNT4, and WNT10B, primarily targeting FZD4_LRP5 and FZD6_LRP5 complexes in both intra- and intercluster communication (Fig. 5E). These interactions displayed high expression magnitude (0.85–0.9) and interaction specificity. Additionally, ECM-related interactions (e.g., LAMB1-ITGA9/ITGB1, VCAN-ITGB1) and EGFR pathway ligands (NRG3, ST6GAL1) further enriched the signaling landscape, suggesting a pro-proliferative and migratory signaling environment. In contrast, zG-like cells in LR1 tumors showed a distinct signaling profile, largely devoid of canonical WNT ligands. Instead, interactions were enriched for cell adhesion and developmental pathways, including NCAM1-FGFR2, HRAS-AGTR1, and LAMA2-ITGA1 (Supplementary Fig. S10, Supplementary Data S6). These interactions point toward a more differentiated and structurally organized state, with lower interaction specificity and more limited signaling breadth compared to HR tumors. Notably, the absence of canonical WNT ligand engagement in LR1, LR2, and LR3 further highlights the molecular divergence between the subgroups.
Fig. 5: Chromatin accessibility and transcription factor motif enrichment define subgroup-specific regulatory programs in pediatric ACT.
A UMAP embedding of snRNA-Seq data from the validation cohort (n = 10 pACT samples, n = 50,988 nuclei), confirming cell type annotations observed in the primary cohort. Transitional cluster exhibiting features of both zG- and zF/zR-like tumor cells is highlighted. B Proportional distribution of metasignatures across DNA methylation–based risk groups in the original and validation cohort (VC). C Chromatin accessibility tracks from snATAC-Seq data showing elevated accessibility at the WNT5A locus (n = 7). D Pseudotime trajectory of tumor cell differentiation inferred from snRNA-Seq and snATAC-Seq integration, showing dynamic expression (right) and activity (left) of selected transcription factors, including TCF7L2 and members of the AP-1 complex (FOS, JUN, JUND) (n = 7). Created in BioRender. Fincke, V. (2026) https://BioRender.com/5kt1mj1. E Motif enrichment analysis based on Tn5 insertion frequency around TCF7L2 binding motifs. High Risk tumors exhibit the strongest enrichment at this CTNNB1-binding motif (n = 7). Equivalent figure for AP-1 is displayed in Supplementary Fig. S15D. F ChromVar scores for CTNNB1-binding and AP-1 binding motifs across risk groups (n = 7). Source data are provided as a Source Data file.G Top enriched transcription factor motifs in High Risk tumors derived from snATAC-Seq, showing selective cso-enrichment of WNT/TCF and AP-1 binding sites (n = 7). H Overview of the mechnismns potentially involved in driving malignancy of the pACT High Risk subgroup. Created in BioRender. Fincke, V. (2026) https://BioRender.com/e38pwvx.
Given the prominent role of WNT signaling, we evaluated CTNNB1 localization by immunohistochemistry as a surrogate marker of pathway activation. HR and LR2 tumors exhibited nuclear accumulation of CTNNB1, whereas LR1 and LR3 tumors showed exclusively membranous staining, supporting subgroup-specific differences in WNT activation (Fig. 4F). All samples in the LR2 and HR subgroup with somatic CTNNB1 mutations exhibited nuclear localization of CTNNB1 (5/5, 100%). Notably, however, 71.5% (5/7) HR tumors without CTNNB1 mutations also showed robust nuclear accumulation, suggesting that WNT pathway activation in this subgroup occurs through alternative mechanisms. Despite this shared feature between HR and LR2, their divergent clinical outcomes indicate that WNT activation alone is insufficient to explain aggressiveness, and that additional cooperating factors likely contribute to the malignant phenotype in HR tumors.
Cell type and metasignature distributions remain consistent in an independent validation cohort, demonstrating reproducibility
To validate the identified cell types in our tumor samples, we performed additional snRNA-seq (n = 3) and combined single-nucleus RNA and ATAC sequencing (snMultiome, n = 7) on additional cases. Filtering and cell-type annotation were performed following the same protocol as in the primary cohort, resulting in 50,799 cells with an average of 2252 detected genes per cell. The identified cell types in the validation cohort closely mirrored those from the original dataset, confirming the reproducibility of our cell type annotations. Notably, we identified a “Transitioning” cluster—exhibiting transcriptional characteristics intermediate between zG cells and zF/zR-like tumor populations – which was predominantly enriched in HR tumors (20.1% in the validation cohort; 0.6–9% in LR groups, Fig. 5A). Importantly, the proportional distributions of tumor and microenvironmental cell types were remarkably consistent between the cohorts. For example, zG-like tumor cells were most abundant in LR1 (36.6% in the original cohort; 28.9% in the validation cohort), while zF/zR-like Cluster II dominated LR3 (61.1% and 85.0%, respectively). These concordant patterns underscore the robustness of our integrated classification and the biological reproducibility of the subgroup-specific cellular architecture (Supplementary Fig. S13, Supplementary Data S7). Previously defined metasignatures were found again and analyzed across subgroups. As shown in Fig. 5B, HR tumors exhibited a highly heterogeneous composition, with all seven metasignatures present and none exceeding ~32% of the transcriptional profile, whereas LR tumors were dominated by a single metasignature.
To further investigate the biological divergence between LR2 and HR tumors—aspects not explained by WNT signaling alone, which is upregulated in both groups—we performed an integrated single-cell multiomic analysis, combining snRNA-seq and snATAC-seq data. Joint UMAP-based dimensionality reduction confirmed consistent separation of the subgroups across transcriptomic, chromatin accessibility, and weighted nearest neighbor (WNN) modalities (Supplementary Fig. S14).
Regulatory motif analysis suggests CTNNB1- and AP1-binding motifs as drivers of tumor biology in high risk pACTs
To further explore the regulatory underpinnings of subgroup-specific biology, we leveraged the snATAC-seq data generated as part of the reproducibility analysis described above. Given that the genetic landscape of the subgroups—with only two samples in the HR group displaying CTNNB1 mutations—does not provide a clear explanation for the overrepresentation of the WNT pathway in the HR group, we hypothesized that epigenetic changes may be responsible for this characteristic. SnATAC-seq revealed increased accessibility at regulatory regions of key WNT pathway genes, including e.g. WNT5A particularly in HR tumors (Fig. 5C). Pseudotime analysis reconstructed a trajectory starting from zG-like tumor cells toward more differentiated zF/zR-like populations (Supplementary Fig. S15A, B). Of note, key effectors of the WNT pathway such as TCF7L2 or LEF were overrepresented early in pseudotime which is consistent with the notion that WNT signaling is a property exerted mainly by un differentiated zG-like cells in the HR group. A further salient pattern in the pseudotime analysis was the overrepresentation of AP-1 transcription factor complex components such as FOS, JUN, and JUND which peaked later in the trajectory (Fig. 5D; Supplementary Fig. S15C).
A detailed motif enrichment analysis of TCF7L2 and FOS::JUND binding sites revealed a striking gradient in chromatin accessibility across risk groups: HR tumors showed the strongest enrichment at FOS::JUND motifs, followed by LR2, whereas LR1 exhibited minimal accessibility at these sites (Supplementary Fig. S15D). Notably, no significant difference in AP-1 binding motif accessibility was observed between HR and LR1, consistent with both groups displaying AP-1 regulatory activity; however, HR tumors uniquely co-activate WNT signaling alongside AP-1 programs, as reflected by concurrent TCF7L2 motif enrichment, whereas LR1 activity appears restricted to AP-1 alone. This accessibility gradient closely parallels the transcriptional activity of AP-1 components observed in snRNA-seq, supporting a model in which HR tumors—unlike the other LR groups—co-activate WNT and AP-1 regulatory programs.
HR tumors demonstrated markedly increased chromatin accessibility and transcription factor (TF) activity involving both TCF7L2 and AP-1 family members, whereas LR subgroups (LR1 and LR2) exhibited selective enrichment for only one TF class, respectively (Fig. 5F). Consistent with this, ChromVAR analysis revealed substantially elevated motif deviation scores in HR tumors compared to LR counterparts, particularly for AP-1 complexes, including FOS::JUND and FOS::JUN Increased TCF7L2 activity was likewise observed in HR tumors, highlighting a distinct transcriptional regulatory program associated with aggressive tumor biology (Fig. 5G). In contrast, LR2 tumors lacked AP-1 motif enrichment, and LR1 tumors showed no evidence of WNT pathway activation, further corroborated by the absence of nuclear CTNNB1 localization (Fig. 5H). Collectively, these data point toward transcriptional cooperativity between WNT and AP-1 effectors as a defining regulatory hallmark of HR pACT, potentially underpinning its more aggressive clinical behavior.
High resolution spatial transcriptomics analysis revealed diverse cell type composition in high risk tumors
Hematoxylin and eosin (H&E) staining reveals typical histomorphological features across all tumors (Supplementary Fig. S16), but spatially resolved transcriptomic data enables the precise localization and annotation of tumor, immune, stromal, and vascular cell types based on previously characterized single-nucleus expression profiles (Fig. 6A–D).
Fig. 6: Spatial and epigenetic reprogramming define high-risk architecture and therapeutic vulnerabilities in pACT and ACC models.
Spatial transcriptomic analysis of one representative tumor per DNA methylation-based subgroup using the Visium HD platform: A Low Risk 1, B Low Risk 2, C Low Risk 3, D High Risk. Cells were reconstructed from 2 µm bins with bin2cell and annotated by transferring the single-nucleus RNA-seq cluster labels with CellTypist. Cells analysed: Low Risk 1, n = 289,651 (tumor compartment 243,148, 83.9%); Low Risk 2, n = 140,105 (108,281, 77.3%); Low Risk 3, n = 99,273 (89,028, 89.7%); High Risk, n = 146,167 (67,396, 46.1%). The stacked bars show the cell type composition across all reconstructed cells (“all”) and across the tumor compartment only (“tumor only”). The Low Risk tumors (A–C) display a compartmentalised architecture, whereas the High Risk tumor (D) shows a disordered structure with an expanded stromal compartment. Insets show PROGENy-inferred WNT signaling activity for the boxed region on an identical color scale across all panels (−2 to 3), comprising 26,579 (A), 30,432 (B), 17,483 (C) and 19,975 (D) cells, respectively; the highest WNT activity was observed in the High Risk section. One section was analysed per subgroup, so no statistical comparison between subgroups was performed and the differences shown are descriptive. E Dose–response viability assays of H295R and CU-ACC1 cells treated with entinostat for 72 h at increasing concentrations. Data are presented as mean values ± SEM (n = 3 independent biological replicates per cell line). F Global methylation changes at the top 1% most variable promoter-associated CpG sites following Entinostat treatment of H295R, showing hypomethylation across conditions (n = 1000 CpG-sites per condition). G UMAP projection of scRNA-seq profiles from treated and untreated ACC cells (H295R and ACC1), demonstrating distinct clustering and transcriptional reprogramming upon HDAC inhibition. (H) Expression heatmap of AP-1 transcription factor components (FOS, FOSB, FOSL2, JUN, JUND, JUNB, ATF2) in treated versus untreated ACC cells, showing consistent downregulation after Entinostat exposure. (I) Bar plots show mean fold change (FC = 2⁻ΔΔCt) in mRNA expression of FOS, FOSB, JUN, and JUNB relative to the 0 min timepoint, in H295R (blue) and ACC1 (red) cells at the indicated timepoints after induction. Individual data points represent biologically independent replicates (n = 3 per group; open circles). Error bars denote ± standard error of the mean (SEM). The dashed horizontal line indicates FC = 1 (no change). Statistical comparisons were performed using one-way ANOVA followed by Tukey’s honestly significant difference (HSD) post-hoc test, comparing each timepoint to the 0 min baseline within each cell line; this test is two-sided and inherently adjusts for multiple comparisons. Asterisks denote Tukey-adjusted p values (*p < 0.05, **p < 0.01, ***p < 0.001); bars not marked with an asterisk were not statistically significant. Values, exact p-values, F-statistics, degrees of freedom, and 95% confidence intervals for all pairwise comparisons are provided in Source Data.
Cell type deconvolution uncovered marked differences in tissue architecture between subgroups: LR tumors exhibited a compartmentalized structure, with well-demarcated tumor and stromal regions and minimal intermixing of immune or fibroblast populations (Fig. 6A–C). In contrast, the HR tumor displayed a disordered architecture, with intermingling of tumor cells with immune and stromal elements and a loss of defined compartment boundaries (Fig. 6D).
To quantitatively assess spatial organization across tumors, we computed Moran’s I spatial autocorrelation for each annotated cell type. LR tumors were largely diffuse, with most populations showing weak to negligible autocorrelation (per-sample mean Moran’s I 0.05–0.08). In contrast, the HR tumor exhibited markedly stronger spatial structuring, with the highest per-sample mean Moran’s I (0.14) and the majority of populations reaching at least moderate autocorrelation. This was most pronounced in fibroblasts, which showed strong spatial clustering in the HR tumor (Moran’s I ≈ 0.34, versus ≤0.20 in LR samples). Among tumor cell types, the Glomerulosa-like population was distinctly more co-localized in the HR tumor (≈0.14 versus ≤0.01 in LR), whereas the zF/zR-like (Cluster II) population was comparably structured across the HR and LR tumors (0.17 in HR vs. 0.08–0.21 in LR). Together, these findings indicate the emergence of localized stromal and partial tumor-cell niches within the HR tumor, consistent with a spatial consolidation of microenvironmental compartments. These spatial differences in cellular organization raised the possibility that regional microenvironmental structure may be accompanied by localized variation in signaling pathway activity. In line with this, AP-1 family members showed highest expression in the HR tumor with elevation in tumor rich regions (Supplementary Fig. S17).
To infer signaling pathway activity across spatial domains, we used the PROGENy compendium to assess pathway-responsive gene signatures. This analysis revealed elevated WNT pathway activity in the HR tumor relative to all LR subgroups, with a modest increase also detected in LR2 (Fig. 6A–D, WNT score insets). These findings align with single-cell and epigenomic evidence of WNT pathway activation in HR tumors.
Together, these findings indicate that HR pACTs are characterized not only by distinct transcriptional and epigenetic states but also by altered spatial tissue organization, including reduced global compartmentalization alongside the emergence of localized stromal and tumor niches with elevated signaling activity. This spatial reconfiguration of the tumor microenvironment may reflect microenvironmental stabilization processes associated with aggressive tumor behavior and potential immune escape.
HDAC inhibition by entinostat reduces viability in vitro, reduces CpG island methylation and alters metabolic programs in ACC cells
Previous analyses identified the HR pACT subgroup as characterized by co-activation of WNT and AP-1 transcriptional programs, alongside an enhanced CpG Island Methylation. To explore targeted therapeutic strategies for this subgroup, we initiated drug screening using a short-term culture derived from a freshly resected metastasis from a patient with HR pACC. Among a panel of candidate compounds (Supplementary Fig. S18, Supplementary Data S8), the class I HDAC inhibitor entinostat showed the most potent anti-proliferative effect, with an IC50 of ~1 µM (Supplementary Fig. S18). In contrast, conventional chemotherapeutic agents such as doxorubicin required substantially higher concentrations to achieve similar levels of viability reduction.
To further dissect the molecular consequences of HDAC inhibition, we treated two adult ACC cell lines (H295R and CU-ACC1) – which share key molecular features with pediatric HR tumors – with entinostat (0.01–50 µM) for 72 h. IC50 values in both lines were comparable to those observed in the short-term culture, confirming consistent drug sensitivity. For comparison, we included decitabine, a DNA-demethylating agent, to assess the relative efficacy of these two epigenetic modifiers. Viability assays confirmed a dose-dependent decrease in cell viability with entinostat treatment in both cell lines, whereas decitabine had minimal effect on viability (Fig. 6E), suggesting that HDAC inhibition may be more effective than DNA demethylation alone in modulating ACC cell growth.
Genome-wide methylation profiling revealed a pronounced reduction in DNA methylation for the 1% most variable promoter-associated CpG sites upon entinostat and dectiabine treatment (Fig. 6F). This hypomethylation was observed across all treatment durations and cell lines (Supplementary Fig. S19A–E). Importantly, targeted methylation analysis for cell line H295R revealed reduced methylation at pro-apoptotic loci including CASP8, CDKN2A, and BCL2L1. CU-ACC1 showed decreased methylation of CDKN2A, APC and CASP9 after entinostat treatment (Supplementary Fig. S19F-G), suggesting potential reactivation of silenced apoptotic pathways.
Dimensionality reduction using UMAP of scRNA-seq data from two cell lines revealed clear transcriptional divergence between entinostat-treated and untreated cells (Fig. 6G), indicating global shifts in gene expression following HDAC inhibition (Supplementary Fig. S20A-B). Analysis of transcription factor expression dynamics showed consistent downregulation of AP-1 components, including FOS, JUND, and JUNB, after treatment (Fig. 6H), aligning with our earlier identification of AP-1 as a central regulatory axis in HR-pACT and suggesting that HDAC inhibition may act through a downregulation of this high risk transcription factor program.
To confirm AP-1 pathway activity in both ACC cell lines, we performed qRT-PCR following serum starvation and re-stimulation in H295R and CU-ACC1 cells. FOS, FOSB, JUN, and JUNB were rapidly and significantly induced in both lines, with induction detectable as early as 15 min and peaking up to 132-fold for FOSB at 120 min in H295R cells (one-way ANOVA, all p < 0.001; Fig. 6I). These results confirm that AP-1 components are stimulus-responsive in both adult ACC cell lines, supporting their validity as models for the AP-1-driven program identified in HR-pACT.
Marker gene heatmaps confirmed a broad reprogramming of transcriptional states in response to entinostat, with loss of several high-risk–associated gene signatures (Supplementary Fig. S20B). Functionally, pathway-level analysis of the scRNA-seq data demonstrated significant upregulation of steroidogenic genes (e.g., STAR, CYP11A1), indicating a shift towards adrenal-specific metabolic programs (Supplementary Fig. S20D-E). To validate these findings in vitro, we performed qRT-PCR for key steroidogenesis genes following entinostat treatment in both cell lines. The response was gene- and cell-line-dependent: in both cell lines, STAR was significantly upregulated (up to 1.6-fold in H295R) while CYP11A1 was dose-dependently downregulated (to 0.81-fold in H295R). Modest but significant upregulation of CYP21A2 was observed at 0.5 µM in both cell lines (Tukey HSD, p < 0.05; Supplementary Fig. S20F). Together, these results indicate that entinostat does upregulate steroidogenic transcription in adult ACC lines.
Additionally, expression of pro-apoptotic genes (including BAX, TNFRSF10B, CASP8, and CDKN2A) was increased in a dose-dependent manner in both cell lines following treatment (Supplementary Fig. S21A-B), in line with the changes observed in the methylome, suggesting enhanced apoptotic susceptibility. Consistent with these changes, Annexin V / 7-AAD flow cytometry analyses revealed an increased fraction of apoptotic cells following entinostat treatment in both ACC models, supporting functional reactivation of apoptotic pathways (Supplementary Fig. S21C).
Together, these single-cell analyses from two ACC models provide evidence that entinostat induces a coordinated epigenetic and transcriptional reprogramming, reversing specific promoter associated methylation, silencing AP-1–driven oncogenic programs, upregulating steroidogenic metabolism, and upregulating pro-apoptotic pathways. Importantly, the functional activity of the AP-1 pathway in both H295R and CU-ACC1 cells was confirmed by qRT-PCR, demonstrating rapid transcriptional induction of FOS, FOSB, JUN, and JUNB upon serum re-stimulation. These findings, together with the results from the short-term culture of a pACC, underscore the therapeutic potential of HDAC inhibition to target a core molecular vulnerability of HR-pACT and may open up a path for the clinical use of these therapeutics in the clinic, probably in concert with other e.g. adrenolytic drugs already used in the treatment of pACT.
