Comprehensive molecular characterization of adenoid cystic carcinoma reveals tumor suppressors as novel drivers and prognostic biomarkers

Adenoidcysticcarcinoma(ACC)isaMYB-drivenheadandneckmalignancywithhighratesoflocalrecurrenceanddistant metastasisandpoorlong-termsurvival.Neweffectivetargetedtherapiesandclinicallyusefulbiomarkersforpatient strati ﬁ cation are needed to improve ACC patient survival. Here, we present an integrated copy number and transcriptomic analysis of ACC to identify novel driver genes and prognostic biomarkers. A total of 598 ACCs were studied. Clinical follow-up was available from 366 patients, the largest cohort analyzed to date. Copy number losses of 1p36 (70/492; 14%) and of the tumor suppressor gene PARK2 (6q26) (85/343; 25%) were prognostic biomarkers; patients with concurrent losses ( n = 20) had signi ﬁ cantly shorter overall survival (OS) than those with one or no deletions ( p < 0.0001). Deletion of 1p36 independently predicted short OS in multivariate analysis ( p = 0.02). Two pro-apoptotic genes, TP73 and KIF1B , were identi ﬁ ed as putative 1p36 tumor suppressor genes whose reduced expression was associated with poor survival and increased resistance to apoptosis. PARK2 expression was markedly reduced in tumors with 6q deletions, and PARK2 knockdown increased spherogenesis and decreased apoptosis, indicating that PARK2 is a tumor suppressor in ACC. Moreover, analysis of the global gene expression pattern in 30 ACCs revealed a transcriptomic signature associated withshort OS,multiple copynumberalterationsincluding 1p36 deletions,and reducedexpressionof TP73 . Taken together, the results indicate that TP73 and PARK2 are novel putative tumor suppressor genes and potential prognostic biomarkers in ACC. Our studies provide new important insights into the pathogenesis of ACC. The results have important implications for biomarker-driven strati ﬁ cation of patients in clinical trials.


Introduction
Adenoid cystic carcinoma (ACC) is an aggressive cancer that predominantly affects the head and neck but also occurs in the breast, tracheobronchial tree, skin, prostate, and female genital tract [1,2].
ACC is generally resistant to therapy, and there are no systemic treatments available for inoperable recurrent or metastatic tumors [3,4].The prognosis for head and neck ACC is poor, and most patients with recurrent/metastatic tumors will die from their disease [1,5,6].
The genomic hallmark of ACC is a translocationgenerated oncogenic gene fusion encoding the N-terminal part of the transcription factor MYB linked to the C-terminal part of the transcription factor NFIB [7][8][9].In a subset of ACCs, MYB is replaced by MYBL1 linked to NFIB or other fusion partners [10][11][12].The MYB::NFIB fusion drives proliferation of ACC cells and is crucial for ACC spherogenesis [13,14].Notably, the fusion is regulated by IGF1R in an AKT-dependent manner [13,15].
In contrast to the near-universal MYB/MYBL1 activation in ACC, the frequency of other mutations in individual genes is low [2,13,[16][17][18][19][20][21][22][23][24].Nonetheless, certain mutations seem to cluster in specific pathways such as NOTCH and IGF/FGF signaling as well as in genes involved in chromatin remodeling.Further progress in understanding the molecular pathogenesis and clinical significance of recurrent genomic alterations in this enigmatic and aggressive cancer will require studies of larger well-characterized patient cohorts with long-term follow-up.Here, we used an integrated approach to study the copy number and transcriptomic profiles of a large cohort of ACC.Clinical follow-up was available from 366 patients, the largest cohort analyzed to date.We identified novel oncogenic drivers, including the tumor suppressor gene PARK2 and pro-apoptotic genes in 1p36 whose reduced expression are associated with aggressive behavior and poor survival in ACC.

Tumor material and patient data
Fresh frozen and formalin-fixed paraffin embedded (FFPE) tumor tissues were available from surgical specimens of 100 head and neck ACCs (supplementary material, Table S1).The genomic profiles of 40 of these have been reported [25].In addition to the 100 ACC specimens, we also analyzed tissue microarrays (TMAs) containing 498 cases, bringing the total number of analyzed ACCs to 598 (supplementary material, Figure S1).
In the TMAs, each tumor was represented by at least two core biopsies.FFPE tissue blocks and TMA sections were obtained from the Departments of Pathology at University Medical Center Hamburg-Eppendorf, University of Texas MD Anderson Cancer Center, University of Alabama at Birmingham, University of Virginia Health System/Charlottesville, Instituto Português de Oncologia Francisco Gentil, and Sahlgrenska University Hospital in Gothenburg, Sweden.Survival data were available for 366 cases.In addition, we had access to the following clinicopathological parameters: sex, age, localization, stage, metastasis, and tumor grade (supplementary material, Table S2).Tumors were graded as lesions with no solid component (grade 1), <30% solid component (grade 2), and ≥30% solid component (grade 3) [26].
Only specimens with >70% tumor cellularity were used for genomic profiling.

Ethical statement
The study was approved by the ethics committee of Instituto Português de Oncologia Francisco Gentil -Lisboa (UIC-1042), the MD Anderson Cancer Center Institutional Review Board (IRB), the University of Virginia IRB, the University of Alabama at Birmingham IRB, and the regional ethics committee in Gothenburg (D-No: 178-08).The use of archived diagnostic leftover tissues for manufacturing of TMAs (Institute for Pathology, University Medical Center Hamburg-Eppendorf, Germany) and their analysis for research purposes was approved by local laws (HmbKHG, §12.1) and by the local ethics committee (Ethics Commission Hamburg, WF-049/09).The work was done in compliance with the Declaration of Helsinki.

Copy number profiling
Genomic DNA was isolated from fresh frozen tumor tissue (n = 58) and FFPE tumor tissue (n = 42) as described [25].High-resolution array comparative genomic hybridization (arrayCGH) was done with the SurePrint G3 Human CGH Microarray 4X180K/244K oligonucleotide arrays (Agilent Technologies, Santa Clara, CA, USA) as recommended by the manufacturer [27].The data were analyzed with Nexus Copy Number software version 8.1 (BioDiscovery, El Segundo, CA, USA) using the FASST2 segmentation algorithm to define nonrandom regions of copy number alterations (CNAs) across the genome.The significance threshold for segmentation was p = 1.0E-6 for fresh frozen tumors and p = 1.0E-8 for FFPE tumors.Similarly, the log2 ratio thresholds for gains and losses were 0.25 and À0.2 for fresh frozen tumors and 0.3 and À0.3 for FFPE tumors, respectively.Each aberration was checked manually to confirm the accuracy of the call.Sex chromosomes and regions partially or completely covered by a copy number variation in the Database of Genomic Variants (http://dgvbeta.tcag.ca/dgv/app/news?ref=NCBI36/hg19) were not analyzed [28].GISTIC2.0 was used to identify genomic regions significantly gained/amplified/deleted [29].The Benjamini-Hochberg method was used to correct for false discovery rate (FDR), and regions with Q values <0.05 were considered significant.CNAs present in 10 or more cases (10%) were considered recurrent.The arrayCGH data are available for download from the Gene Expression Omnibus (GEO) database (https://ncbi.nlm.nih.gov/geo/,Accession No. GSE153228).and PARK2 (6q26) (PARK2 FISH probe from Empire Genomics (Buffalo, NY, USA) and a subtelomeric 6qter probe from Leica Biosystems, Nussloch, Germany), as previously described [24,27].The protocols for pretreatment, hybridization, and posthybridization washes were as recommended by the manufacturers.Cell nuclei were stained with DAPI.Fluorescence signals were digitized, processed, and analyzed with the CytoVision image-analysis system (Applied Imaging, San Jose, CA, USA) and Isis FISH imaging system version 5.5 (MetaSystems, Altlussheim, Germany).At least 20 nuclei were scored from each core biopsy.

Sphere, proliferation, adhesion, and apoptosis assays
Cultured ACC cells were derived from ACC67 (grade 1) [13] and ACC100 (grade 3) (supplementary material, Table S1).Both tumors were MYB::NFIB fusion positive.In addition, ACC67 showed gain of 19p13.11-pteras the only copy number change, whereas ACC100 had 17 different gains and losses, including loss of 1p36 and PARK2 (6q26).Low-passage, mycoplasmafree cells were used and validated for MYB::NFIB fusion transcripts, as described [13].Cells were transfected with 50 nM Silencer Select siRNAs for KIF1B (s23022, s23024), MFN2 (s19261, s19262), PARK2 (s10043, s10044), TP73 (s14320, s14321), or negative control siRNAs using the Lipofectamine RNAiMAX transfection reagent (Thermo Fisher Scientific) [13].For sphere assays, cells were grown for 10 days under nonadherent conditions in six-well plates precoated with PolyHEMA (Merck, Darmstadt, Germany), as described [35].For proliferation assays, two million cells were seeded in T25 flasks and counted 7 days later using a Countess 3 cell automated counter (Thermo Fisher Scientific).Alternatively, 400,000 cells were seeded in six-well plates and grown for 48 h.Cells were supplemented with 10 μM BrdU during the last 16 h and subsequently fixed and stained using a FITC BrdU Flow kit (Becton, Dickinson and Company, BD, Franklin Lakes, NJ, USA).Stained cells were analyzed with the Accuri C6 flow cytometer (BD).Cell adhesion was evaluated in 96-well plates precoated with collagen type I solution (Merck).Twenty thousand cells were allowed to adhere overnight and then treated with 0.25% Trypsin-EDTA solution (Thermo Fisher Scientific) for 2.5 min.The relative number of remaining adherent cells was estimated with the Alamar Blue reagent (Thermo Fisher Scientific).For apoptosis assays, cells were seeded in 96-well plates (BD) and treated with 50 nM siRNAs for 96 h followed by 1 μM doxorubicin, vinorelbine, palbociclib, or AZD4573 (Selleck Chemicals, Houston, TX, USA) for 24 h and analyzed with the Caspase-Glo 3/7 reagent (Promega, Madison, WI, USA) [14].Cellular protein expression was analyzed by immunoblotting using mouse monoclonal antibodies to PARK2 (mAb 4211, Cell Signaling Technology, Danvers, MA, USA) and beta-actin (ab8226, Abcam, Cambridge, UK) [13].Live tumor cells were imaged with a Zeiss Axiovert A1 microscope equipped with an AxioCam ERc5s camera and cell size was measured using a Countess 3 cell automated counter (Thermo Fisher Scientific).

Statistical analyses
Associations between CNAs and clinicopathological parameters as well as the fraction of downregulated genes within the 1p36 deletion compared with the copy number neutral chromosome 13 were analyzed using a χ 2 test.Overall survival (OS) was defined as the time from initial diagnosis to death (any cause) or last follow-up.Survival was estimated using Kaplan-Meier tests, and differences between groups were analyzed with the log-rank test.
Genomic imbalances with breakpoints in or immediately distal to MYB were detected in 16% of ACCs.Two tumors had interstitial deletions within 1 Mb telomeric to MYB, and one had an interstitial deletion 0.59 Mb centromeric to MYB.Four tumors showed gain of one MYB allele (supplementary material, Figure S5).
Eighteen focal amplifications were identified in six tumors (supplementary material, Table S5 and Figure S5).Amplifications of KIT and PDGFRA were identified in two cases, high-level amplification of MYB in one case, and amplifications of HMGA2 and MDM2 in one.Homozygous deletion of CDKN2A/B and a chromothrypsis-like event in 12p11.23-q23.3 were seen in one case each (supplementary material, Table S5).
PARK2 is a putative tumor suppressor gene in ACC GISTIC2.0 identified a peak region in 6q25.3-q26,containing PARK2 as the only gene (Figure 1B,D).RT-qPCR analysis of 30 ACCs revealed that PARK2 expression was significantly lower in tumors with 6q deletions than in those without (Figure 1G).ACC cells with 6q/PARK2 deletion (ACC100) showed significantly higher sphere formation compared with ACC cells without deletion (ACC67) (Figure 2A).To investigate whether decreased levels of PARK2 affected the spherogenesis of ACC cells, we knocked down PARK2 expression in cultured ACC cells without 6q deletion.RT-qPCR and western blot analyses revealed robust knockdown of PARK2 mRNA and protein levels in ACC67 cells (Figure 2B), whereas PARK2 mRNA and protein expression were not repeatedly detected or were undetected in ACC100 cells.PARK2 downregulation led to a two-to three-fold increase in the formation of ACC tumor spheres under anchorage-independent growth conditions (Figure 2C).Notably, ACC cells with PARK2 deletion showed decreased apoptosis in response to doxorubicin treatment compared to cells without PARK2 deletion (Figure 2D).This is further supported by studies showing that PARK2 knockdown leads to decreased apoptosis in response to doxorubicin treatment (Figure 2E).Taken together, our findings suggest that PARK2 has tumor suppressor activity with effects on both tumor growth and resistance to therapy in ACC.

Molecular consequences of 1p36 deletion
To study the molecular consequences of 1p36 deletions, we analyzed the mRNA expression of genes located within 1p36.32-p36.21(GISTIC2.0region) in 30 tumors with (n = 10) and without (n = 20) 1p36 deletion.Twenty-seven of the 128 genes in this segment were significantly downregulated in tumors with 1p36 deletion ( p < 0.05) (Figure 2F) (supplementary material, Table S6), which is significantly more than expected (p < 0.0001).These results were confirmed by GSEA (Figure 2G).In a gene ontology analysis, apoptosis and Novel drivers and prognostic biomarkers in adenoid cystic carcinoma the pentose-phosphate shunt were significantly enriched biological processes associated with the downregulated genes in 1p36 (Figure 2H).RT-qPCR analysis confirmed that the pro-apoptotic genes TP73, MFN2, KIF1B, DFFA, and DFFB in 1p36 were significantly downregulated in tumors with 1p36 deletion (Figure 2I), suggesting that the copy number losses in 1p36 are associated with resistance to apoptosis.To test this hypothesis, we transfected cultured ACC cells from two cases (ACC67 and ACC100) with individual siRNAs targeting the early pro-apoptotic genes TP73, MFN2, and KIF1B for 5 days (Figure 2J).We also   Novel drivers and prognostic biomarkers in adenoid cystic carcinoma 261 treated siRNA-transfected cells with doxorubicin or vinorelbine during the last 24 h and measured caspase 3/7 activity.Downregulation of any of the three genes significantly decreased apoptosis induced by these chemotherapeutic drugs, though the effect was weaker for MFN2 and only significant with one siRNA for each case (Figure 2K; supplementary material, Figure S6A-C).Thus, deletion of TP73 and KIF1B promotes resistance to apoptosis in ACC cells.
Next, we compared the mutation frequency of genes in 1p36.32-p36.21 with that of all genes in 83 sequenced ACCs in The Cancer Genome Atlas (TCGA) [16,17].We found no significant difference ( p = 0.55); only four nonrecurrent mutations were identified in genes in this region, and only one tumor had a 1p36 deletion.Evidently, 1p36 deletions result in a gene dosage effect and downregulation of several pro-apoptotic genes, rendering ACCs with 1p36 deletions resistant to apoptosis.

Deletions of 1p36 and PARK2 correlate with survival and tumor grade
To further validate the significance of 1p36 and PARK2 deletions, we studied their frequency by FISH in an independent large cohort of ACCs.Deletions of 1p36 were found in 13.5% of the cases (53/392 analyzable cases showed loss of one or both 1p36 probe signals) and loss of one PARK2 allele in 24.3% (59/243 analyzable cases) (Figure 3A,B).
To evaluate the clinical significance of the 1p36 and PARK2 deletions, we studied the correlation between these aberrations and various clinicopathological parameters in our ACC cohort (supplementary material, Table S7).Survival analyses revealed significantly shorter OS in patients with 1p36 deletions than in those without (p < 0.0001) (Figure 4A); patients with 1p36 deletions also had shorter median survival (38 versus 147 months; 95% CI 34-42 and 110-184, respectively) and shorter 5-year OS (27% versus 77%).Patients with PARK2 deletions also had shorter OS than those without (p = 0.0008) (Figure 4B).Notably, patients with concurrent deletions of 1p36 and PARK2 had shorter OS than patients with either one or neither of these deletions (p < 0.0001) (Figure 4C).
Next, we estimated the hazard function of death of ACC patients with Cox regression.In univariate analysis, the hazard function was significantly dependent on 1p36 deletion, PARK2 deletion, age at diagnosis, stage III and IV, and tumor grade 3 (supplementary material, Table S8).In multivariate analysis, the significant risk factors of death were 1p36 deletion, age, stage III and IV, and tumor grade 3. Median survival of patients with or without 1p36 deletion calculated with the hazard function also revealed that the effect of 1p36 deletion on survival is dependent on the age of the patient at diagnosis (Figure 4D).At 53 years of age (the median age at diagnosis in our study), a patient with a 1p36 deletion has an expected survival of less than 5 years versus $15 years in a patient without the deletion.
Deletions of 1p36 were overrepresented in tumors with solid histology ( p < 0.0001) (Figure 4E), as were deletions of PARK2 (p = 0.002) (data not shown) and deletions of both 1p36 and PARK2 (p < 0.0001) (Figure 4F).Cox regression analysis of grade 3 tumors only (n = 63) revealed that 1p36 deletion (p = 0.01) and PARK2 deletion (p < 0.001) predicted shorter OS in this select subgroup.Furthermore, the risk of death in patients having 1p36 deletion was similar between different tumor grades, whereas it was increased in grade 3 tumors with PARK2 deletion ( p = 0.005).These results indicate that 1p36 deletion predicts shorter OS regardless of tumor grade, whereas PARK2 deletion confers an additional risk in high-grade tumors.Collectively, these findings show that 1p36 and PARK2 deletions are recurrent prognostic biomarkers in ACC and that 1p36 deletion is an independent predictor of OS in multivariate analysis.

A gene expression signature that defines a subset of ACC with 1p36 deletion and short OS
To study the effect of genomic alterations on gene expression, we performed unsupervised hierarchical clustering and principal component analysis of global gene expression data from 30 ACCs with known genomic profiles.The ACCs formed two major clustersdesignated ACC-I and ACC-IIthat were clearly separated from NSG (Figure 5A,B).MYB or MYBL1 was activated in all tumors in both clusters (supplementary material, Figure S7).
Tumors in the ACC-I cluster had markedly more CNAs than those in the ACC-II cluster (Figure 5C).Novel drivers and prognostic biomarkers in adenoid cystic carcinoma 263

M Persson, MK Andersson et al
Deletions of 1p36 were found exclusively in ACC-I tumors, and PARK2 deletions were more frequent in ACC-I tumors (Figure 5A,C).Gains of 6p and 8q were found only in ACC-I tumors (Figure 5C).In contrast, all but one of the tumors with 12q deletions were in the ACC-II cluster.ACC-I tumors, including four lacking 1p36 deletions, had significantly lower expression of the pro-apoptotic gene TP73 than ACC-II tumors (Figures 2I and 5D).Importantly, OS was markedly shorter in patients with ACC-I tumors (Figure 5E).Thus, the gene expression and copy number profiles are significantly associated with the prognosis of ACC.
To investigate the biological processes associated with the expression profiles of the two clusters, we used GSEA.Gene sets downregulated in invasive and metastatic tumors were significantly associated with ACC-I tumors (Figure 5F).In global gene expression and gene ontology analyses, the most significantly downregulated genes (163 genes, p < 0.0001) in the ACC-I versus ACC-II cluster (Figure 5G) were involved in cell adhesion and extracellular matrix organization (Figure 5H).Expression of genes involved in cell proliferation (e.g., MKI67; p = 0.002) was higher in ACC-I than in ACC-II tumors (Figure 5I), suggesting that ACC-I tumors grow faster and are more prone to metastasize than ACC-II tumors.
Next, we studied cellular characteristics of cultured ACC cells with (ACC100) and without (ACC67) 1p36 and PARK2 deletions.ACC cells with 1p36/PARK2 deletions had a higher proliferation rate and reduced surface adhesion compared to cells without deletions (supplementary material, Figure S8A-C).Moreover, ACC cells with 1p36/PARK2 deletions had a smaller size, grew in multilayers, and had a less prominent cytoplasm (supplementary material, Figure S8D,E).Both ACC cells with and without 1p36/PARK2 deletions were resistant to treatment with the CDK4/6 inhibitor palbociclib (supplementary material, Figure S8F).However, cells with these deletions were sensitive to the CDK9 inhibitor AZD4573 (supplementary material, Figure S8G), and knockdown of TP73, KIF1B, and PARK2 reduced the apoptotic response induced by this drug (supplementary material, Figure S8H).

Discussion
A major challenge in the management of ACC patients is the high rate of recurrences and distant metastases and the lack of effective systemic therapies and clinically useful prognostic biomarkers.Using an integrated genomic and transcriptomic approach, we now demonstrate that deletions of 1p36 and PARK2 are significant prognostic biomarkers in ACC and that loss of pro-apoptotic genes in 1p36 and loss of the tumor suppressor PARK2 contribute to aggressive behavior and poor survival in ACC.
Our analyses identified losses involving 1p36, 6q (PARK2), 9p, and 12q as significant focal CNAs in ACC.OS was significantly shorter in patients with 1p36 or PARK2 deletions than in those without such deletions.This observation was further validated by FISH in a large cohort of ACC patients with clinical follow-up.The 1p36 deletion independently predicted short OS on multivariate analysis consistent with the results of smaller studies of ACC [36,37].Interestingly, we also found that the effect of the 1p36 deletion on survival was dependent on patient age at diagnosis.Finally, we showed that concurrent deletions involving 1p36 and PARK2 were associated with the shortest OS and preferentially occurred in poorly differentiated (grade 3) ACCs.We anticipate that these prognostic biomarkers will be clinically useful for treatment selection and for stratification of patients in clinical trials and that in a clinical setting they are best analyzed by FISH using robust 1p36 and PARK2 probes.Future studies will show whether immunohistochemistry for PARK2 and/or tumor suppressor proteins encoded by genes in 1p36 (e.g., TP73) will be feasible.
The clinical consequences of 1p36 deletions in ACC prompted us to investigate their molecular underpinnings.Gene expression profiling revealed downregulation of a significant number of genes in 1p36 in ACCs with deletions.Further analysis identified apoptosis as a significantly inhibited pathway in tumors with 1p36 deletions.RT-qPCR confirmed downregulation of several pro-apoptotic genes, including the early genes TP73, KIF1B, and MFN2 [38][39][40][41].Consistent with previous observations in other tumors [42], we found evidence of monoallelic expression of TP73 in ACC since tumors with loss of one copy of 1p36 had much lower, or almost undetectable, levels of TP73 than tumors without deletions.Reduced expression of pro-apoptotic genes in 1p36 as a result of haploinsufficiency may lead to inhibition of stress-induced apoptosis and resistance to cytotoxic therapies.Indeed, knockdown of in particular TP73 and KIF1B in ACC cells without 1p36 deletions increased resistance to apoptosis induced by cytostatic drugs.In summary, we identified the pro-apoptotic genes TP73 and KIF1B as potential 1p36 tumor suppressor genes in ACC.Reduced expression of these genes as shown by copy number loss was associated with poor survival and increased resistance to apoptosis.Previous studies of ACC failed to detect any recurrent mutations in genes in 1p36 [16,23], supporting the notion that the major molecular consequences of the 1p36 deletions are dosage-dependent impairment of multiple genes in this region.This conclusion is also consistent with studies of other cancer types with recurrent copy number losses in 1p36 [43][44][45].
Copy number profiling identified the tumor suppressor gene PARK2 [46][47][48] as a single candidate target gene of the 6q deletions in ACC.Expression analysis confirmed that PARK2 expression is indeed lower in ACCs with 6q26 deletions than in those without.Moreover, PARK2 knockdown in ACC cells without 6q26 deletions increased spherogenesis and decreased apoptosis in response to chemotherapeutic drugs.Thus, PARK2 loss increases tumor growth initiated by ACC stem/progenitor cells and increases resistance to apoptosis.PARK2 encodes the E3 ubiquitin ligase Parkin Novel drivers and prognostic biomarkers in adenoid cystic carcinoma [49,50] and is an important cell-cycle regulator and a master regulator of G1/S cyclins [51].PARK2 also regulates apoptosis through BCL-XL [52], and loss of PARK2 function can activate PI3K/AKT signaling through inactivation of PTEN [53].The PARK2 tumor suppressor is mutated, or more commonly deleted and downregulated, in for example, lung cancer, glioblastoma, colon cancer, ovarian cancer, and melanoma [46][47][48][54][55][56].Our findings suggest that PARK2 is also a tumor suppressor in ACC that is associated with poor prognosis in a subset of tumors.This observation is further supported by our recent study showing that losses of sequences distal to the MYB locus in 6q23.3 preferentially occur in grade 3 tumors and are associated with poor OS in ACC [24].
Integrated transcriptomic and copy number profiling revealed several previously unrecognized events associated with ACC tumorigenesis.Unsupervised cluster analysis identified two major clusters of ACCs that differed in genomic profile and clinical outcome.Tumors in the ACC-I cluster had 1p36 deletions, multiple CNAs, and gains of 6p and 8q, whereas tumors in the ACC-II cluster lacked these aberrations, had fewer CNAs, and preferentially had 12q deletions.Although present in both clusters, PARK2 deletions were more frequent in the ACC-I cluster.Importantly, patients in this cluster also had a significantly shorter OS.Similar groups with different expression profiles and prognosis have previously been identified in transcriptomic studies of ACCs [57,58].In the present study, the transcriptomic profiles of ACC-I tumors were associated with increased cell proliferation, invasion, and metastasis, consistent with the aggressiveness of these tumors.Collectively, our results identified a gene expression signature associated with short OS, multiple CNAs including 1p36 deletions, and reduced expression of TP73.
A limitation of our study is the paucity of functional studies of the candidate tumor suppressor genes in 1p36 and distal 6q, mainly reflecting the lack of authentic ACC cell lines and the difficulty of culturing ACC cells in vitro [13,59].Another limitation is that extensive mutational data were not available for our patient cohort.On the other hand, the comprehensive genomic data from 366 ACC patients with clinical follow-up enabled us to identify reliable prognostic biomarkers for ACC.
In summary, we used an integrated copy number and transcriptomic approach to identify novel driver genes and prognostic biomarkers in ACC.We found that copy number losses of 1p36 and PARK2 are potential prognostic biomarkers in ACC and that loss of pro-apoptotic genes in 1p36 and loss of the tumor suppressor PARK2 contribute to aggressive behavior and short OS in ACC.Our findings provide new important insights into the pathogenesis of ACC that have implications for biomarker-driven stratification of patients in clinical trials.

Figure 2 .
Figure 2. Molecular and functional consequences of PARK2 and 1p36 deletions in ACC.(A) Spherogenesis in ACC100 cells with PARK2 and 1p36 deletions and in ACC67 cells without these deletions.Cells were cultured under nonadherent conditions for 10 days.Data show the average of three independent experiments.(B) RT-qPCR and western blot analyses of PARK2 mRNA and protein expression in ACC cells after treatment with 50 nM PARK2 siRNAs for 48 h.(C) Sphere formation 10 days after PARK2 knockdown in ACC cells.Experiments were done three times with six biological replicates.(D) ACC100 cells are less sensitive to doxorubicin treatment (1 μM, 24 h) compared to ACC67 cells.(E) Knockdown of PARK2 decreases apoptosis (caspase 3/7 activity) in cultured ACC cells after 24 h treatment with 1 μM doxorubicin.Data represent one of three independent experiments with five biological replicates.(F) Distribution of 1p deletions (red) detected by arrayCGH in 100 ACCs.The lower panel shows the expression of differentially expressed genes in tumors with and without 1p deletions.Underexpressed genes are indicated in red.(G) GSEA showing enrichment of downregulated genes in a publicly available 1p36 gene set in ACCs with 1p36 deletion.(H) Gene Ontology analysis of genes downregulated in 1p36 in tumors with 1p36 deletion.(I) RT-qPCR analysis of pro-apoptotic genes located in 1p36 in ACCs with (n = 10) or without (n = 20) 1p36 deletion (Mann-Whitney test).(J) Knockdown of pro-apoptotic genes TP73, MFN2, and KIF1B in ACC67 and ACC100 cells.Data show one of three independent experiments with three biological replicates.(K) Knockdown of 1p36 genes decreases apoptosis (caspase 3/7 activity) in cultured ACC cells after treatment with doxorubicin.Results shown represent one of three independent experiments with five biological replicates.Mean ± SEM. **p < 0.01, ***p < 0.001, ****p < 0.0001 by independent sample t-test or one-way ANOVA.

Figure 3 .
Figure 3. FISH analyses of 1p36 and PARK2 gene deletions in ACC.(A) ACC with deletion in 1p36, shown by loss of one KLHDC7A (1p36.13)red signal and retention of two green signals from the centromere probe.Arrowheads indicate retained KLHDC7A allele.(B) ACC with loss of one copy each of PARK2 (6q26) and 6q subtelomeric region, shown by loss of one red and one green signal.Arrowheads indicate remaining intact copy of PARK2 and 6q subtelomeric region (red/green signals).Nuclei are stained in blue with DAPI.Ideograms of chromosomes 1 and 6 with the location of the FISH probes used are shown to the left of each FISH image.

Figure 4 .
Figure 4. OS of ACC patients with 1p36 and/or PARK2 deletions.(A) 1p36 deletion versus no 1p36 deletion (log-rank test).(B) PARK2 deletion versus no PARK2 deletion (log-rank test).(C) 1p36 and PARK2 deletions versus no 1p36 and PARK2 deletions (log-rank test).(D) The effect of 1p36 deletion on survival is significantly dependent on patient age at diagnosis.Dashed lines indicate median survival for 53-year-old ACC patients (median age of the patients in this investigation) with and without 1p36 deletion.(E and F) Pie charts show significant overrepresentation of 1p36 deletions (E) and of concurrent 1p36 and PARK2 deletions (F) in grade 3 ACCs (χ 2 test).

Figure 5 .
Figure 5. Global gene expression analysis identifies two clusters of ACC with significantly different copy number profiles and patient survival.(A) Unsupervised hierarchical clustering and heatmap analysis of global gene expression in 30 ACCs and 7 NSGs.The presence or absence of deletions of 1p36, PARK2, and 12q and the MYB/MYBL1 status are indicated.(B) Principal component analysis of global gene expression in 30 ACCs and seven NSGs. (C) CNAs in tumors in ACC-I and ACC-II clusters.(D) RT-qPCR analysis of 30 ACC surgical specimens showing significantly lower TP73 expression in ACC-I tumors than in ACC-II tumors.Mean ± SEM (Mann-Whitney test).(E) Kaplan-Meier survival analysis of patients in ACC-I and ACC-II clusters (log-rank test).(F) GSEA of tumors with and without 1p36 deletion.(G) Volcano plot showing most significantly upregulated (red) and downregulated (blue) genes in ACC-I versus ACC-II cluster.(H) Gene Ontology analysis of genes downregulated in ACC-I versus ACC-II tumors.(I) GSEA of global gene expression showing enrichment of Hallmark gene sets in ACC-I versus ACC-II tumors.
The hazard function of death was estimated by Cox regression.Multivariate analyses were adjusted for 1p36 deletion, age at diagnosis, stage, and tumor grade.Confirmation of the proportional hazards assumption and linearity of Martingale and deviance residuals was done with the survival package in R. Median survival as a function of 1p36 deletion and age at diagnosis was estimated with Poisson regression.Differences in gene expression in patient samples measured by RT-qPCR were evaluated using nonparametric Mann-Whitney tests.
Significant differences between groups were assessed by one-way ANOVA or independent samples t-test.Prism version 9 (Graphpad, La Jolla, CA, USA), 258 M Persson, MK Andersson et al SAS version 9.3 (SAS Institute, Cary, NC, USA), SPSS Statistics version 28 (IBM, Armonk, NY, USA), and R (https://www.r-project.org)were used for statistical analyses; p < 0.05 was considered statistically significant.