Alternative lengthening of telomeres in childhood neuroblastoma from genome to proteome

Telomere maintenance by telomerase activation or alternative lengthening of telomeres (ALT) is a major determinant of poor outcome in neuroblastoma. Here, we screen for ALT in primary and relapsed neuroblastomas (n = 760) and characterize its features using multi-omics profiling. ALT-positive tumors are molecularly distinct from other neuroblastoma subtypes and enriched in a population-based clinical sequencing study cohort for relapsed cases. They display reduced ATRX/DAXX complex abundance, due to either ATRX mutations (55%) or low protein expression. The heterochromatic histone mark H3K9me3 recognized by ATRX is enriched at the telomeres of ALT-positive tumors. Notably, we find a high frequency of telomeric repeat loci with a neuroblastoma ALT-specific hotspot on chr1q42.2 and loss of the adjacent chromosomal segment forming a neo-telomere. ALT-positive neuroblastomas proliferate slowly, which is reflected by a protracted clinical course of disease. Nevertheless, children with an ALT-positive neuroblastoma have dismal outcome.

A ctivation of a telomere maintenance mechanism (TMM) was found to be associated with poor outcome in neuroblastoma 1 , which is the most common extracranial solid tumor diagnosed in early childhood [2][3][4][5] . Telomerase activation is one way of maintaining telomere length and results from amplified MYCN (MYCN transcriptionally activates the TERT gene) or TERT rearrangement in neuroblastoma 1,6 . Activation of alternative lengthening of telomeres (ALT) presents an alternative route of telomere maintenance, which is seen in a substantial proportion of neuroblastoma tumors 1,[6][7][8] . Mutations in the chromatin modifier alpha thalassemia/mental retardation syndrome X-linked (ATRX) are associated with elongated telomeres and ALT in neuroblastoma 7,8 . Furthermore, deletions of ATRX were associated with recurrent partial chromosomal losses in neuroblastomas without amplified MYCN 9 . ATRX and its complex partner death associated protein 6 (DAXX) incorporate histone variant H3.3 at pericentric and telomeric repeat sequences [10][11][12] . ALT activity is strongly associated with somatic ATRX mutations in different tumor entities [13][14][15][16] . Although loss of ATRX alone does not induce ALT, ATRX acts as an ALT suppressor and overexpression of ATRX in ATRX-mutant cells represses ALT markers 17,18 . Interestingly, knockdown of DAXX abolishes the ALT suppressive effect of ATRX overexpression in ATRX-mutant cells highlighting the importance of both complex members 17,18 .
The clinical neuroblastoma risk stratification systems used by most national and international consortia (Society For Pediatric Oncology and Hematology-GPOH, International Society of Paediatric Oncology Europe Neuroblastoma Group-SIOPEN, Children's Oncology Group-COG), as well as the International Neuroblastoma Risk Group (INRG) staging system, categorizes neuroblastomas mainly based on INSS stage, age, and MYCN status into low risk, intermediate risk, and high risk 19 . Thus in clinical practice, ALT-positive high-risk tumors are treated with the same treatment protocol as MYCN-amplified and TERTrearranged tumors presuming similar biological features as these telomerase-activated tumors 1 .
Focused studies on ALT in neuroblastoma are rare and previous observations were mostly based on genetic sequencing studies and ATRX mutation status as ALT marker [6][7][8] . Here, we aim to deepen the knowledge about the biology and clinical features of ALT-positive neuroblastomas by enriching ALT tumors in the study cohort independent of ATRX mutation status. Using a multi-omics profiling approach of ALT-positive neuroblastomas, we provide evidence that this subgroup is clinically and molecularly distinct. Clinically, ALT is associated with a protracted course of disease and poor outcome. Molecularly, ALT-positive tumors feature mutated ATRX and/or reduced protein abundance, heterochromatic telomeric chromatin, a slow proliferative capacity, and frequent neo-telomere formation.

Results
Enrichment of ALT in relapsed disease. Using the presence of circular partially single stranded extrachromosomal telomeric repeat sequences (C-Circles) as ALT marker 20 , 9.2% of the neuroblastomas in a screening cohort of 720 tumors were classified as ALT-positive. C-Circle screening was further extended to a relapse cohort of tumors analyzed in the INFORM registry 21 (n = 40). Importantly, 47.5% of the INFORM tumors were ALTactivated, indicating a strong enrichment of this molecular subgroup in relapsed cases (Fig. 1a, Tables 1-2). Clinically, ALT was significantly associated with INSS stage 4, high-risk disease, and an older age at diagnosis. However, about one third of the ALTpositive neuroblastomas had localized disease and were stratified at diagnosis as low or intermediate risk using a standard risk classification system. In the screening cohort, none of the 83 tested 4 S stage tumors, a clinically recognizable subtype associated with spontaneous regression seen in infants below one year of age, were ALT-activated. Amplified MYCN and C-Circle presence were mutually exclusive in the screening cohort, but not in the INFORM cohort (Tables 1-2

, Supplementary Data 1-2).
Association of ALT with protracted clinical disease course and poor outcome. Subsequently, neuroblastomas enriched for ALT were characterized using a multi-omics approach including high coverage whole genome sequencing (n = 165), RNA-sequencing (n = 144), whole proteome analysis (n = 34), and ChIPsequencing (n = 27) (Fig. 1b, Supplementary Data 3, Methods) forming the discovery cohort. ALT tumors had a significantly higher telomere content (Fig. 2a). Telomere content and C-Circle intensity was comparable between ALT-positive tumors of the discovery and INFORM cohort ( Supplementary Figs. 1-2). Overall, telomere content was positively correlated with C-Circle intensity (r = 0.76, P < 2.2e −16 , Spearman, Supplementary  Fig. 1c). Minimal to no TERT mRNA expression was observed for ALT-positive neuroblastomas (Fig. 2b), resulting from epigenetic silencing of the TERT locus by H3K27me3 ( Supplementary  Fig. 3a). In accordance with low TERT mRNA expression, ALTpositive tumors exhibited low telomerase activity ( Supplementary  Fig. 3b). Moreover, ALT was associated with increased expression of the polyadenylated telomeric long noncoding RNA TERRA (Fig. 2c, Supplementary Fig. 3c). Patients with ALT-positive tumors had a similar poor event-free survival compared to patients with MYCN-amplified tumors. Although the overall survival time of patients with ALT-positive tumors was significantly longer, they had a dismal outcome in the long perspective (Fig. 2d). No difference in event-free or overall survival was observed between patients with ALT-positive tumors stratified into high-risk and those predicted as low/intermediate risk (Fig. 2e), indicating that the current risk stratification is underestimating the poor prognosis of ALT-positive tumors. ALTpositive neuroblastomas with a higher telomere content had a significantly shorter event-free survival as compared to ALTpositive tumors with relatively lower telomere content (Supplementary Fig. 3d). There was no clear trend in telomere content gain or loss between matching primary/relapse samples of the INFORM cohort ( Supplementary Fig. 4). For 3 of 7 ALT-positive relapse tumors with a matching primary, ALT activity was only observed in the relapse tumor. Two of these cases were MYCN amplified in the primary disease period and heterogeneous ALT/ MNA in the relapse. Heterogeneity could be confirmed by FISH analysis (MYCN and telomere probe) of one relapsed tumor ( Supplementary Fig. 5a). For the other heterogeneous relapse tumor, we found evidence of an ALT-positive subclone in FISH analysis of the primary tumor ( Supplementary Fig. 5b). Noteworthy, the discovery cohort also contains three heterogeneous ALT/MNA cases, which are MYCN amplified and have a very high telomere content, but were otherwise C-Circle negative (Fig. 2a, Supplementary Figs. 1a and 5c).
ATRX mutations in 55% of ALT-positive neuroblastomas. The mutational landscape of ALT-positive neuroblastomas in the discovery cohort was very diverse with only few recurrent genetic events (Fig. 3a, Supplementary Fig. 6). ALT activation was mutually exclusive to TERT rearrangements (Fig. 3a). Loss-offunction mutations of ATRX were the most frequent events associated with ALT, but were only observed in 55% of ALTpositive tumors (Fig. 3a). ATRX mutations were mainly large deletions and single nucleotide variants ( Fig. 3a-b, Supplementary Data 3-4). However, three patients had large intragenic ATRX duplications ( Fig. 3a-b, Supplementary Data 3-4). Patient survival was not significantly different between ATRX-mutated and ATRX wild-type ALT-positive cases (Fig. 3c). Mutations in the ATRX complex members DAXX and H3F3A were extremely rare (Fig. 3a). Only one ATRX wild-type tumor exhibited a large inversion event affecting DAXX, associated with low DAXX mRNA expression (66.46 TPM; range discovery cohort 50-330 TPM). Another tumor had an A92S H3F3A mutation in addition to mutated ATRX. Somatic mutations in TP53 pathway genes (TP53, CREBBP, ATM, ATR, CDKN2A, and MDM2) were significantly enriched (P = 0.01) in ALT-positive tumors. ALTpositive tumors had the highest prevalence of CDK4 amplifications (3/4) (Fig. 3a), associated with increased CDK4 mRNA expression ( Supplementary Fig. 7a), and showed higher CCND1 mRNA expression as compared to telomerase-activated cases ( Supplementary Fig. 7b). Overall, many tumors had mutations in one or another gene associated with telomere maintenance according to the TelNet database 22 (Fig. 3a, Supplementary  Figs. 6 and 7c). Copy number loss of POLD3 and ATM was frequently observed in ALT-positive tumors ( Supplementary   Fig. 7c). However, both POLD3 and ATM are located on chr11q and loss of these genes was associated with loss of chr11q, which is frequently observed in ALT-positive ( Supplementary Fig. 8) and other non-MYCN-amplified neuroblastomas 23,24 . Synaptic nuclear envelope protein 1 (SYNE1) was exclusively mutated in 7 of 60 ALT tumors (11.7%) (Fig. 3a, Supplementary Figs. 6, 9a). Furthermore, deletions in receptor-type tyrosine-protein phosphatase delta (PTPRD) were enriched in ALT-positive tumors (P = 2.46e-05; Fig. 3a, Supplementary Fig. 6). Structural variations affecting PTPRD were reported in neuroblastoma cohorts and low PTPRD expression was correlated with a poor prognosis 25,26 . However, structural variations of PTPRD were not correlated with reduced PTPRD mRNA expression in our discovery cohort ( Supplementary Fig. 9b). Noteworthy, both genes are very large and thus the high mutation frequency in ALT neuroblastomas could also reflect the general increased mutational load found in older patients or a general increase in genomic instability ( Supplementary Fig. 9c- 1 C-Circle screening of neuroblastomas. a ALT frequency detected using C-Circle presence as a marker in the screening cohort (n = 720, left) and INFORM cohort (n = 40, right). Top pies illustrate the composition of the cohorts concerning INSS tumor stage. b Schematic overview of the analyzed patient cohorts used for C-Circle screening and high-throughput analysis using whole genome sequencing (WGS), RNA-sequencing, whole proteome analysis, and ChIP-sequencing. Sample numbers are given in brackets. Number of C-Circle positive (CC+) and C-Circle negative (CC-) patients in the subcohorts are indicated. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21247-8 ARTICLE comparable to the discovery cohort ( Supplementary Fig. 10a, Supplementary Data 2, 3). In one matched primary relapse pair, an ATRX duplication event was retained in the relapse, further indicating that duplications can lead to nonfunctional ATRX ( Supplementary Fig. 4). Notably, canonical activating RAS pathway mutations (HRAS, NRAS, KRAS, BRAF, RAF1, CDK4, CCND1, NF1) were significantly more frequent in relapsed ALTpositive neuroblastomas in the INFORM cohort compared to the discovery cohort (P = 0.0013, Supplementary Fig. 10, Fig. 3a), supporting a specific impact of RAS pathway mutations in relapsed ALT-positive tumors.
Low ATRX protein abundance in ALT tumors. Using a mass spectrometry based whole proteome approach (n = 34) quantifying on average 6891 proteins per tumor, we found 470 proteins significantly different in ALT-positive tumors compared to the other subgroups (nominal P ≤ 0.01; fold change ≥2; Fig. 4a, Supplementary Data 5). Enrichment analysis of significantly different mRNAs and proteins for each subgroup in turn highlighted both the overall agreement of transcriptomics and proteomics as well as the distinct effects only significantly present in one or the other omics-level (Fig. 4b, Supplementary Fig. 11a, Supplementary Data 6). The Ras family proteins HRAS and NRAS, comprised in the term "prenylation", were significantly upregulated in ALT-positive neuroblastomas (Fig. 4b, Supplementary Data 6). Among the top upregulated proteins in ALT-positive tumors were the cancer testis antigens P antigen family member 5 (PAGE5) and melanoma associated antigen 4 (MAGEA4), which may serve as potential targets for immunotherapies (Fig. 4a). Many pathways associated with proliferation were significantly downregulated on protein level in ALT-positive tumors compared to the telomerase-activated tumors including "DNA replication" and "chromosome". Notably, the classical proliferation marker MKI67 27 , comprised in the term "chromosome", was one of the top significantly downregulated proteins in ALT-positive Table 1 Clinical features of C-Circle positive tumors in the screening cohort (n = 720) a .  neuroblastomas ( Fig. 4a-b, Supplementary Fig. 11b). The term DNA replication includes among others the proteins PCNA, MCM2, MCM3, MCM4, MCM5, and MCM7, which are also indicative of the proliferative capacity of a cell 28 . Furthermore, the term "bromodomain" was significantly reduced in ALTpositive tumors, while strongly upregulated in MNA tumors. Amplified MYCN is known to be associated with increased occupancy of active promoter regions and enhancer invasion by MYCN and bromodomain proteins, leading to an increased transcriptional activity of many proliferation associated genes and downregulation of differentiation genes 29 . Moreover, ALTpositive tumors exhibit a significantly lower fraction of cycling cells compared to MYCN-amplified tumors ( Supplementary  Fig. 11c). Taken together, both protein expression data and cell  cycle analysis support a low proliferative capacity of ALT-positive neuroblastomas. 96 of the 470 significantly different proteins were annotated in the TelNet database to be associated with telomere maintenance 22 (Supplementary Data 5) (P = 0.03). The significantly downregulated proteins in ALT-positive neuroblastomas comprised ATRX, PCNA, EXO1, GNL3, KDM1A, RIF1, SCG2, TFAP2B, and GNL3L, all functionally linked to telomere maintenance 22 . Helicases and chromatin regulators, including ATRX, were less abundant at protein level in the ALT subgroup, but not at mRNA level (Fig. 4b). Importantly, ATRX itself was among the top candidates to exhibit lower protein abundance in ALT-positive tumors (Fig. 4a). Intriguingly, the reduced ATRX protein levels were independent of ATRX mRNA levels and mutation status (Fig. 5a). ATRX was among the most strongly downregulated proteins while the ATRX mRNA changes very little ( Supplementary Fig. 11d), indicating that reduced ATRX protein abundance is a characteristic proteomic feature of all ALT-positive tumors (ATRX wild-type and mutated) that cannot be observed at the mRNA level. Analysis of exon-specific ATRX mRNA levels revealed that only the exons affected by deletion have reduced expression in the ATRX-deleted cases ( Supplementary Fig. 12a). In summary, reduced ATRX protein level is a biomarker for ALT-positive neuroblastomas that is independent of mRNA levels and ATRX mutation. The decreased ATRX protein levels despite unchanged mRNA abundance could be due to reduced translation and/or increased degradation of ATRX in ALT-positive tumors. For example, it has been shown that unassembled subunits of multiprotein complexes (so-called orphans) are often degraded 30 . Downregulating one subunit can therefore induce degradation of other subunits, which partially explains divergence of protein-and mRNA-level changes in many biological contexts 31,32 . We therefore took a closer look at the ATRX binding partner DAXX. Interestingly, DAXX protein levels were specifically reduced in ALT-positive ATRX wild-type tumors (Fig. 5b). To investigate if reduced DAXX levels could explain ATRX downregulation at the protein level, we knocked-down DAXX in the ALT-negative neuroblastoma cell line NBL-S and observed that ATRX protein levels were indeed reduced after 96 h ( Fig. 5c-d). Hence, the reduced ATRX protein levels in ATRX wild-type ALT positive tumors might result from downregulation of the DAXX protein. However, since DAXX mRNA levels do not differ significantly between tumors (Fig. 5b) and no recurrent mutation patterns in DAXX or ATRX/DAXX interacting proteins could be identified in ALT-positive ATRX wild-type tumors (Supplementary Figs. 12b,13), the question what causes the downregulation of DAXX is still open. Hence, the mechanistic details behind ATRX/DAXX complex reduction remain to be uncovered.

Heterochromatic telomeric chromatin in ALT neuroblastomas.
Many contradictory findings were reported on the telomeric chromatin landscape of ALT-positive cells [33][34][35] , but data on primary tumor samples was yet missing. Here, ALT-positive neuroblastoma tumors were found to be enriched for H3K9me3 at the telomeres (Fig. 6a). Overall levels of H3K9me3 at the telomeres of ALT-positive tumors were comparable to the H3K9me3 levels at SatII and SatIII alpha-satellite sequences (Supplementary Fig. 14a-b), which are known to be enriched for H3K9me3 35,36 . In line with a heterochromatic enrichment, protein expression of the demethylase KDM1A responsible for removing methyl groups from H3K9me3 37 was significantly lower in ALT-positive neuroblastomas compared to TERT-activated tumors (Fig. 6b). SETDB1 protein activity is required for H3K9me3 deposition at the telomeres, while the histone methyltransferases SUV39H1 and SUV39H2 catalyze H3K9 trimethylation at heterochromatic sites [38][39][40] . Neither expression of SETDB1 nor SUV39H1/SUV39H2 could be associated with the increased telomeric H3K9me3 in ALT-positive neuroblastomas ( Fig. 6c-d). H3K27ac at the telomeres was lower in ALT-positive tumors (Fig. 6a). However, this difference was not telomere specific and similarly observed at SatII and SatIII sequences indicating a general difference in H3K27ac ( Supplementary  Fig. 14c). Furthermore, ALT-positive tumors displayed reduced telomeric H3K27me3 (Fig. 6a, Supplementary Fig. 14d) and low protein and mRNA expression of the PRC2 component EZH2, responsible for H3K27me3 deposition (Fig. 6e). Low H3K36me3 was observed at the telomeres and satellite sequences of both ALT-positive and ALT-negative neuroblastomas (Fig. 6a, Supplementary Fig. 14e). Together, this indicates epigenetic dysregulation of repressive marks at the telomeres in ALT-positive tumors with excess of H3K9me3 and reduction of H3K27me3.
Increased frequency of telomeric repeat loci in ALT tumors. ALT-positive neuroblastomas exhibited an increased frequency of telomeric repeat loci compared to the other groups ( Fig. 7a), with ALT-positive tumors displaying up to 19 such events per tumor. The telomeric repeat loci frequency was significantly higher in ATRX-mutated than ATRX wild-type ALT tumors. Telomeric repeat loci are characterized by a chromosomal junction site and the presence of telomeric repeat sequences either upstream or downstream of this junction site (one-sided) (Fig. 7b). In rare cases, two telomeric repeat loci in close proximity (10 kb) with opposite orientation can form a two-sided event ( Supplementary  Fig. 15a). In our discovery cohort, only five such two-sided events were observed ( Supplementary Fig. 15b). Two-sided events can result either from two separate events in close proximity or from telomere sequences that are spanning the region between the two junction sites forming a true insertion. Evidence for a true insertion, meaning that mates of a read pair matched to both sides of the insertion could only be identified for two of these events (Supplementary Table 1). The presence of more than three telomeric repeat loci in a tumor was found to be associated with a significantly reduced overall survival in ALT-positive neuroblastomas (Fig. 7c). Hotspot regions for recurrent telomeric Fig. 2 Characteristics and survival of ALT-positive neuroblastomas. a Relative telomere content (tumor/control) of neuroblastoma tumors in the discovery cohort grouped according to the genetic evidence of an active telomere maintenance mechanism (see methods for details): ALT activation (ALT, red), telomerase activation by amplified MYCN (MNA, yellow), structural variation affecting TERT or TERT promoter mutation (TERT, blue), and patients having no genetic evidence of an activated telomere maintenance mechanism (OTHER, gray). b TERT mRNA expression in different subgroups. c Expression of the telomeric repeat-containing RNA TERRA in the different subgroups. a-c Boxplots indicate the median value (middle line), the 25th and 75th percentiles (box). The upper/lower whisker spans from the hinge to the largest/smallest value (values expanding a distance of 1.5 x inter quartile range are not considered). Dots represent individual tumors. P values were calculated with two-sided Wilcoxon rank sum tests. d Event-free and overall survival probability compared between ALT-positive (n = 60) and MYCN-amplified (MNA, n = 34) neuroblastomas. e Event-free and overall survival probability of patients with ALT-positive tumors classified as low or intermediate risk (LR/IR, n = 20) compared to patients with ALT-positive tumors classified as high-risk (HR, n = 39) using the NB2004 risk stratification. d-e P values were calculated using log rank tests. a-e n describes the number of analyzed tumors.
repeat loci were found on chromosomes 1q42.2, 18q23, and 19q13.43 (Fig. 7d). The increased occurrence of telomeric repeat loci in ALT-positive neuroblastomas and their enrichment on 1q42.2 was confirmed in the INFORM dataset ( Supplementary  Fig. 15c). 80.3% (212/264) of all t-type (TTAGGG) containing telomeric repeat loci showed at least 1 bp microhomology to the canonical telomeric repeat sequence, which is more than one would expect by chance. This enhanced microhomology at the junction sites of telomeric repeat loci indicates that telomeric sequences were added via a microhomology-dependent mechanism ( Supplementary Fig. 15d-f). In general, 55.3% of all telomeric repeat loci were associated with a chromosomal copy b 76 Fig. 15g). A smaller fraction of loci (4.9%) was associated with a copy number gain. However, 30.7% of the telomeric repeat loci were copy number neutral. Some of these copy number neutral events were associated with one or more structural variations, but for 77.2% of the copy number neutral telomeric repeat loci no other event could be detected ( Supplementary Fig. 15g).  (Fig. 7g). Among the significantly reduced genes in 1q42.2-1qter deleted tumors were the genes ARID4B, EXO1, FH, CNST, ADSS, HNRNPU, and RYR2, which are known to be associated with telomere maintenance and biology 22 . Of note, expression of the 5′−3′ exonuclease, EXO1, located on chromosome 1q43 and involved in DNA double strand break repair 41,42 , telomere replication, and t-loop formation 43 , was not only reduced in 1q42.2-deleted tumors but generally reduced in ALT-positive neuroblastomas ( Supplementary Fig. 16c).

Discussion
In summary, this in-depth analysis revealed that ALT-positive neuroblastomas are biologically and clinically distinct from tumors with telomerase activation. ALT in neuroblastoma was associated with characteristic ALT features, including high telomere content, low TERT expression, and increased TERRA expression, which is in line with previous findings on ALT-positive tumors 44,45 . ATRX mutations were identified in 55% of the ALT-positive neuroblastomas. In line with this observation, an ATRX mutation frequency of 60% was reported in a parallel study 46 . The most frequent type of ATRX mutations were large deletions. However, we could also identify focal intragenic ATRX duplication events. Focal duplication of ATRX was also observed in a matching primary/relapse pair indicating that these events are leading to loss of ATRX. Similar events resulted in loss of ATRX function in patients with the genetic disorder alpha thalassemia X-linked intellectual disability syndrome (ATRX-syndrome) 47,48 further supporting a loss of ATRX function by partial duplications.
Integrating proteomic profiling revealed reduced ATRX/DAXX protein complex abundance as a recurrent event in ALT-positive neuroblastoma, which could often not be explained by mutations in these genes. The observation that all five ALT-positive tumors with wild-type ATRX depicted reduced DAXX protein levels is intriguing. It is tempting to speculate that the ALT phenotype always results from loss of the ATRX/DAXX complex activity, which is caused either by ATRX mutations or by reduced ATRX/DAXX protein levels. Future studies will show if this is indeed the case in neuroblastoma and/or possibly other ALT-positive cancers. Nevertheless, we could identify a subgroup of ALT-positive neuroblastomas with wild-type ATRX, but low ATRX protein abundance. Reduced ATRX protein levels in ALT ATRX wildtype neuroblastoma could be explained by the reduced DAXX protein levels, which we specifically observed in this subgroup of tumors. In this scenario, reduced DAXX protein levels impair assembly of the ATRX/DAXX complex, which then causes degradation of orphan ATRX protein molecules. Consistent with this idea, we observed that knocking down DAXX reduced ATRX protein levels in neuroblastoma cells. However, the cause of reduced DAXX protein levels is not yet clear, especially since mRNA levels do not change significantly. Reduced DAXX protein levels in ALT-positive tumors may result from posttranscriptional events. It is known that DAXX is regulated via various posttranslational modifications including phosphorylation, SUMOylation, and ubiquitination 49 . Irrespective of the mechanistic details involved, our study highlights that proteomic data is closer to phenotypes than transcriptomic data, which is especially valuable in a clinical context 32 .
A recent report stated that ATRX mutation and amplified MYCN are mutually exclusive 50 . C-Circle screening of a cohort of 720 neuroblastomas also found a mutual exclusivity of C-Circle presence and amplified MYCN. However, in the INFORM cohort we identified two cases with amplified MYCN and a positive C-Circle signal. Additionally, in our discovery cohort we classified three cases as ALT/MNA heterogeneous. These cases were negative in the C-Circle assay, but exhibited a very high telomere content. FISH analysis of some of these cases could also confirm that amplified MYCN and ultra-bright telomere spots are present in the same tumor ( Supplementary Fig. 5). No ATRX mutations were detected in the MNA/ALT heterogeneous tumors (for NBI26 only lcWGS and WES data was available). Taken together, heterogeneous tumors with amplified MYCN and presence of ALT markers seem to occur, but are very rare. Further analysis and larger sets of these cases would be necessary to draw meaningful conclusions on the clonality of these events and the activity of the respective telomere maintenance mechanism.
ChIP-sequencing analysis of ALT-positive neuroblastoma tumors showed that telomeres as well as SatII and SatIII sequences show lower H3K27ac compared to ALT-negative tumors. Furthermore, proteins comprised in the term "bromodomain", including EP300 responsible for H3K27ac, were reduced in ALT tumors and enhanced in telomerase-activated tumors. Moreover, the telomeres of ALT-positive neuroblastoma tumors were enriched for H3K9me3. These observations confirm the results of a previous study on cell lines 35 that only the telomeres of ALT-positive tumors are enriched for H3K9me3. In addition, we observed low H3K27me3 at the telomeres of ALT cases, which goes in line with a significantly  ALT-positive neuroblastoma tumors exhibited a high rate of telomeric repeat loci. Since telomeric repeat loci were characterized by telomeric repeat sequences either upstream or downstream of a non-telomeric junction site, these events cannot have occurred from breakage of interstitial telomeric sequences (ITS). However, telomeric repeat loci frequently overlapped with breakpoints of copy number changes or structural variations. Because terminal chromosomal breaks are in need of telomeric repeats to protect the newly formed ends from degradation 51-53 , we propose that ALTpositive tumors are capable of adding telomeric repeat sequences to open ends of chromosomal breaks forming neo-telomeres (Fig. 8a). The high degree of microhomology to the telomeric repeat sequence at the junction sites indicates that telomeric repeats are added via a microhomology-dependent process like microhomology-mediated end-joining or non-homologous endjoining 54 . Further, we propose that the presence of microhomology at an open chromosomal break determines if telomeric sequences can be added to this site. Chromosomal loss of certain fragments might present a selection advantage leading to a selection of cells harboring the neo-telomere. We also identified a subset of copy number neutral telomeric repeat loci with no associated structural variation. This might be due to the fact that these events are subclonal and were thus not detected by the CNV/SV calling algorithm. Moreover, the detection limit of the used copy number algorithm is 50 kb and thus smaller copy number changes cannot be detected. Two-sided events, defined as two telomeric repeat loci in a 10 kb window with opposite orientation of the telomeric repeats, were very rare. These events may represent insertions of telomeric repeat sequences similar to previously described events 55 . Only two of five two-sided events exhibited evidence of a true insertion by mates of a read pair mapping to both sides of the insertion. However, for large insertions the used short read sequencing prevents the detection of supporting reads. Alternatively, two-sided events may result from neo-telomere formation on both sides of a breakpoint (Fig. 8b). A recurrent hotspot for telomeric repeat loci was identified on chr1q42, which was associated with 1q42.2-1qter loss forming a neo-telomere on chr1q42.2. Telomeric repeat loci were also associated with ALT and overlapped with chromosomal breakpoints in a pan-cancer cohort, but no enrichment was seen for telomeric repeat loci on 1q42.2 45 , indicating that neo-telomere formation on chromosome 1q42.2 is neuroblastoma-specific. 1q42.2-1qter deletions retained in relapsed ALT cases indicate a selective advantage of the deletion. Furthermore, chr1q42.2-1qter deletions were associated with reduced expression of candidate genes involved in telomere biology, including EXO1. Interestingly, EXO1 expression is not only reduced in ALT-positive tumors with 1q42.2-1qter deletions, but in ALT tumors in general. This indicates a potential role of EXO1 as an ALT suppressor that may be silenced by loss of 1q42.2 or other means. ALT activity was associated with unfavorable outcome and a protracted clinical course of disease, which further substantiates previous observations 56 7 Telomeric repeat loci. a Telomeric repeat loci frequency in ATRX-mutated (dark pink) and ATRX wild-type (light pink) ALT-positive tumors compared to TERT (blue), MNA (MNA), and tumors without genetic evidence of an activated telomere maintenance mechanism (OTHER, gray) in the discovery cohort. Boxplots indicate the median value (middle line) and the 25th and 75th percentiles (box). The upper/lower whisker spans from the hinge to the largest/smallest value (values expanding a distance of 1.5 x inter quartile range are not considered). Dots represent individual tumors. n describes the number of analyzed tumors. P values were calculated with two-sided Wilcoxon rank sum tests. b Example of one telomeric repeat locus in tumor NBD119. Dark blue color indicates sequencing coverage and light blue color illustrates clipped sequences. Reads are shown in medium gray and clipped bases are color-coded. Non-telomeric ends of a discordant read pair are labeled in dark gray. c Overall survival probability of ALT-positive tumors harboring more than three telomeric repeat loci compared to ALT-positive tumors with three or less than three events. P value calculated using a log rank test.   21 . For a subset of these (n = 31) high coverage whole genome sequencing was done. A detailed overview of all clinical parameters and available data of the INFORM cohort is given in Supplementary Data 2. All patients or their legal guardian signed a written informed consent. Genomic DNA was isolated from fresh frozen material using a phenol chloroform extraction and control DNA was isolated from whole blood using the NucleoSpin Blood DNA extraction kit (Macherey-Nagel).
C-Circle assay. C-Circle assays were performed based on a previously described protocol 20 , which is briefly described in the following. 30 ng of genomic DNA were used for every tumor sample. Prior to amplification, tumor DNA was digested with RsaI (4U/ug) and HinfI (4U/ug) and RNAse (25 ng/μg) treated for 1 h at 37°C. 0.2 mg/mL BSA, 0.1% Tween 20, 1 mM dATP, dGTP and dTTP, 2 x Φ29 buffer and 7.5 U Φ29 polymerase was added to the pretreated DNA and rolling circle amplification was done at 30°C for 8 h, followed by an inactivation at 65°C for 20 min. For every sample, a sample without polymerase was included as control and on every blot the neuroblastoma cell lines CHLA-90 (ALT-positive) and SK-N-BE2 (ALT-negative) were included as positive and negative assay controls. Tel-oTTAGGG kit (Roche) was used for hybridization and development according to the manufacturers' instructions. All blots were quantified using the ImageJ gel quantification tool (ImageJ version 1.51). A C-Circle positive signal was defined as minimum 4-fold increase of the area under the curve relative to control without polymerase. In addition, the signal intensity had to be at least 20% of the CHLA-90 positive control. C-Circle status was assessed by quantification and optical inspection of all blots by two investigators. Sequencing libraries for the Illumina HiSeq 2000 platform were generated using the NEB Next Ultra DNA kit (NEB) according to the manufacturers' instructions. Libraries for sequencing on the patterned flowcell v2.5 were generated using the Truseq DNA Nano kit (Illumina) according to the manufacturers' instructions. Libraries were size selected using SPRI beads (Beckman Coulter Genomics). One Touch Pipeline (OTP) service of the German Cancer Research Center (DKFZ) 59 was used for alignment and variant calling. Whole genome sequencing data was aligned to the 1000 Genomes project phase 3 assembly with decoy and PhiX contigs reference genome using BWA-MEM version 0.7.8, 0.7.8-r2.08 or 0.7.15 with option -T 0. Sambamba version 0.6.5 was used for merging and duplication marking and bam files were filtered using Samtools version 0. Genes were defined as amplified if the copy number was at least 8-fold higher than the total copy number of the tumor sample. Any copy number losses that were not homozygous deletions were defined as a deletion. Any loss of heterozygosity (LOH) that was not also a deletion (e.g., copy-neutral LOH) was defined as LOH. Small insertions and deletions were determined based on an OTP in-house workflow 61 -consecutive instances of the four most  common telomeric repeat types (TTAGGG, TCAGGG, TGAGGG, and TTGGGG) were extracted. For the further analysis, only unmapped reads or reads with a very low alignment confidence (mapping quality lower than 8) were used. Telomere read count was normalized by all reads in the sample with a GC-content of 48-52% and furthermore telomere content of tumor samples was normalized to the matching blood control. For tumor samples without a matching control, the mean telomere content of all controls sequenced with the same protocol in the same sequencing center was used for normalization.
Subgrouping based on telomere maintenance mechanism (TMM group). All tumors were categorized into four groups according to the genetic evidence of an activated telomere maintenance mechanism (ALT, MNA, TERT) or the absence of such a mechanism (OTHER). Tumors being either C-Circle positive or having a log2 telomere content relative to the control sample >1 were classified as ALT.
Tumors with an amplified MYCN gene were considered to use telomerase activation by MYCN (MNA). The TERT group comprises tumors with a structural variation affecting TERT or TERT promoter mutation. Tumors with a relative log2 telomere content above one and amplified MYCN were included in the heterogeneous group HET ALT/MNA despite being C-Circle negative.
Telomeric repeat loci. To find telomeric repeat loci we searched for tumor-specific discordant paired-end reads, where one end was an extracted telomere read and the other end was non-telomeric and uniquely mapped to a chromosome (mapping quality >30). In 1 kb regions containing at least four discordant reads in the tumor sample and none in the matching control, exact junction sites were defined by at least three split reads. The split reads had to contain at least one TTAGGG repeat.
Regions with discordant read pairs in at least 15 control samples were excluded. Finally, the junction sites were visualized using the Integrative Genomics Viewer (IGV) 62 Matching primary of NBI26. Amplification status of MYCN was determined as described previously 75 with MYCN Vysis LSI N-MYC (2p24) SpectrumGreen/ CEP2 SpectumOrange and Vysis LSI N-MYC Spectrum orange. Activation of ALT was determined by identification of ALT-associated promyelocytic leukemia (PML) nuclear bodies (APB) by immunofluorescence in combination with FISH using Alexa-488-(TTAGGG)n PNA probes as described previously 6 . Images were acquired using Leica DM 5550B fluorescent light microscope system at 63x resolution. Full z-stacks were taken at 0.5 µm and projected using Cytovision software (Version 7.6).
NBI5. Tumor touch imprints were fixed and labeled as previously described 76  Statistics. Differences between various neuroblastoma subgroups in terms of telomere content, telomeric repeat loci, enrichment of ChIP signals, TPM, and LFQ expression values were tested with two-sided Wilcoxon rank-sum tests using R (R Foundation for Statistical Computing, version 3.5.1 or 3.6.1). All boxplots were generated using the function geom_boxplot() in the R package ggplot2 (version 3.2.0 or 3.3.1). Thus, all boxplots indicate the median value. The lower and upper hinge of the boxplots correspond to the 25th and 75th percentiles. The upper/lower whisker spans from the hinge to the largest/smallest value (values expanding a distance of 1.5 x inter quartile range are not considered). Additionally, individual samples are shown as jitter points. Dotplot showing quantification of ATRX protein levels after DAXX knockdown based on western blot images was done using the function geom_dotplot() in the R package ggplot2 (version 3.3.1). Horizontal line is representing the median value. P values for Western blot quantification were calculated using a two-sided unpaired Student's t-test in R with the function t.test. P values for differential RNA expression analysis were calculated with Wald tests within DESeq2 version 1.18.1 69 . P values for differential protein expression analysis were calculated with two-sided Student's t-test statistics using Perseus software version 1.5.5.3 82 . Correlation analysis was performed using ggpubr (version 0.3.0) or with the function cor in R. Enrichment analysis was tested by hypergeometric test on gene symbols of significantly different proteins and mRNA determined for this analysis by SAM-modified 83 (S0 = 0.01) Welch t-test were performed using R. Multiple testing correction by Benjamini-Hochberg method was performed with the R function p.adjust. Survival curves were generated using Kaplan Meier analysis in the R package survminer (version 0.4.7). P values for survival analysis were calculated using log rank tests. The optimal cut point to assess the prognostic impact of the number of telomeric repeat loci and telomere content was determined using Maximally Selected Rank Statistics with the criteria that at least 20% of the cases have to be in each group using the R package survminer (version 0.4.7). Associations of C-Circle presence with clinical features like age, sex, stage, risk classification was calculated using Fisher tests. Enrichment of TP53 pathway mutations (TP53, CREBBP, ATM, ATR, CDKN2A, and MDM2) and PTPRD mutations was calculated using a Fisher test comparing the occurrence in ALTpositive compared to ALT-negative tumors. Enrichment of RAS pathway mutations (HRAS, KRAS, NRAS, BRAF, RAF1, NF1, CDK4, CCND1) was calculated using a Fisher test comparing their occurrence in ALT-positive tumors in the discovery cohort compared to ALT-positive tumors in the INFORM cohort. HET cases and relapse cases of the discovery cohort were excluded. A Fisher test was used to calculate the enrichment of proteins mentioned in the Telnet database in the differentially expressed proteins of ALT-positive tumors. Fisher tests were done using R (version 3.6.1).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Whole genome sequencing, whole exome sequencing, ChIP-sequencing, RNAsequencing, and proteome data has been archived at the European Genome-phenome Archive [https://www.ebi.ac.uk/ega/] and can be found under accession number EGAS00001004349. Part of the whole genome and RNA-sequencing data was previously published 6 and is available under the accession number EGAS00001001308. The data is deposited under controlled access. Data access can be obtained by contacting Frank Westermann for EGAS00001004349 or Matthias Fischer for EGAS00001001308. Access to data requires a data transfer agreement. mRNA and protein expression data are available as datasets in the R2 database and can be interactively analyzed. mRNA expression data can be found under the name "Tumor Neuroblastoma ALT -Westermann − 144 -tpm -gencode19" [https://hgserver1.amc.nl/cgi-bin/r2/main.cgi? table=ps_avgpres_2010fwr144_gencode19]. Protein data can be found under the name "Tumor Neuoblastoma ALT (Protein) -Westermann -34 -LFQ -fw2010prot" [https:// hgserver1.amc.nl/cgi-bin/r2/main.cgi?table=ps_avgpres_fw2010prot34_fw2010prot]. All functional SNVs, INDELs, SVs and copy number alterations (only amplification and homozygous deletions) for all tumors in the discovery cohort are provided as Supplementary Data 7-9. The remaining data are available within the Article, Supplementary Information or available from the authors upon request. Source data are provided with this paper.