Telomere Maintenance Mechanisms Define Clinical Outcome in High-Risk Neuroblastoma

Unlimited proliferation of cancer cells requires that they maintain telomeres (1), which are typically maintained by activating telomerase (2). TERT encodes the catalytic component of telomerase (3, 4). Cancer cells activate telomerase by acquiring TERT promoter mutations, TERT genomic alterations, or transcriptional dysregulation (5–8). TERT mRNA levels correlate with telomerase activity (9, 10) and is often used to infer telomerase activity in cancer (5, 9–11). Some cancers do not express TERT mRNA and use the recombination-mediated alternative lengthening of telomeres (ALT) mechanism (12). Telomeric repeats in the form of partially double-stranded DNA, termed C-circles, provide a highly specific marker for ALT activity (13). ALT is commonly associated with loss-of-function genomic alterations in ATRX or its binding partner DAXX (14–16), and less commonly with mutations in H3F3A (17) or SMARCAL1 (18).

Neuroblastoma is a malignant childhood cancer arising from the developing sympathetic nervous system with clinical behavior that ranges from spontaneous regression to relentless and fatal tumor progression in spite of multimodal intensive therapy (19). High-risk neuroblastoma, as defined by stage, age, histology, and MYCN genomic amplification, has an overall survival (OS) rate of approximately 50% with current intensive therapy (20–22). In addition to known oncogenic drivers, including MYCN amplification and ALK mutations (23, 24), recent whole-genome and -exome sequencing studies demonstrated structural rearrangements near the TERT locus in approximately 25% of high-risk neuroblastomas (5, 11) and large scale structural variations (SV) in the chromatin remodeler gene ATRX in approximately 10% of high-risk neuroblastomas (24).

Of these four most frequent genomic alterations identified in neuroblastomas, three converge on activating telomere maintenance mechanisms (TMM; refs. 5, 24). Tumors with MYCN amplification or TERT rearrangements activate telomerase by inducing TERT mRNA expression (5, 11), while loss of ATRX is implicated in the ALT phenotype (14–16). Genomic amplification of MYCN, TERT genomic rearrangements, and loss of ATRX are nearly always mutually exclusive events, suggesting a functional redundancy between these mechanistic pathways. Telomere maintenance has long been thought to be pivotal for high-risk neuroblastoma (25). However, neuroblastoma cell lines with high telomere content (TC) and continually shortening telomeres due to lack of a TMM [termed the ever-shorter telomere (EST) phenotype] have been described and analysis of primary patient tumors suggested that EST comprises >10% of high-risk patient tumors (26).

A recent study of TMM in neuroblastoma reported that 3% of high-risk neuroblastoma tumors lacked evidence of TMM activation (27), but that study employed TERT expression by microarray and defined ALT by detecting ALT-associated promyelocytic leukemia bodies (APB). Unlike the specific and sensitive C-circle assay (13), APBs can also be detected in non-ALT cells with long telomeres because of an increased frequency of telomere trimming events (28). Thus, the true frequency and clinical relevance of TMM-negative (EST phenotype) tumors in high-risk neuroblastoma remains to be defined.

To comprehensively define TMMs and their clinical impact in high-risk neuroblastoma, we employed the ALT-specific C-circle assay, TC measurement by qPCR, and TERT mRNA expression measurement by RNA-sequencing (RNA-seq) in 134 neuroblastoma primary tumors (110 high-risk and 24 nonhigh-risk). Genomic sequencing data for most of these tumors were available from the NCI TARGET program (https://ocg.cancer.gov/programs/target; accession no. phs000467) and was integrated with clinical risk stratification and outcome data. We also assessed TERT expression, C-circles, and TC in patient-derived cell lines (n = 103) and xenografts (n = 28) established from high-risk neuroblastomas.

Primary tumor samples prior to therapy were collected (snap-frozen) with written informed consent from children with neuroblastoma entered on to Children's Oncology Group (COG) protocols. Pilot sections of tumors verified tumor content. RNA and DNA were extracted, RNA quality monitored by RNA integrity number (RIN) values, and subjected to RNA and DNA sequencing by the NCI TARGET program (24). DNA and RNA were provided to Texas Tech University Health Sciences Center (TTUHSC, Lubbock, TX) for C-circle assay, TC assay, and RT-PCR for TERT mRNA expression.

SVs from whole genomes (CGI) used a combination of alignment-based and read-depth–based approaches. Alignment-based SVs were initially obtained through The CGI Cancer Pipeline 2.0. Read-depth breakpoints were obtained from segmentation profiles generated from tumor/blood-matched log R ratios from every 2 Kb window across the genome; ATRX variants were as previously described for the whole-exome sequencing (WES) dataset (24, 29).

C-circle assay (13), amplified C-circles, detected using telomere qPCR or slot blotting (26, 30), and TC assay by qPCR (26) were performed as described previously.

TERT mRNA expression was measured by RNA-seq, validated by RT-PCR, details in Supplementary Materials and Methods. Telomeric repeat amplification protocol (TRAP) assay was performed on whole-cell lysate from selected cell lines and patient-derived xenografts (PDX) using either TRAPeze Telomerase Detection Kit (EMD Millipore) or TeloTAGGG Telomerase PCR ELISA Kit (Roche) as described previously (31).

Terminal restriction fragment (TRF) analysis was performed as described previously (32).

For constitutive overexpression, MYCN was cloned into pLenti-C-Myc-DDK-IRES-Puro (OriGene; PS100069).

All neuroblastoma patient-derived cell lines and PDXs were obtained from the ALSF/COG Childhood Cancer Repository at TTUHSC (www.CCcells.org). All cell line and PDX identities were confirmed using the AmpFLSTR Identifier Plus PCR Amplification Kit (Applied Biosystems) at time of use for experimentation, verified against the short tandem repeat database at www.CCCells.org. Neuroblastoma cell lines were used at passage numbers below 25, except when noted in results, were cultured in antibiotic-free medium, and routinely tested for lack of Mycoplasma contamination. A total of 28 neuroblastoma PDXs were used in this study all employed at passage 2 in mice. PDXs were all verified to be free of human pathogens and Epstein–Barr virus by PCR.

R (Version 3.3.2), SAS Software (SAS Institute), IBM SPSS (V25), and GraphPad Prism (version 7.0a) were used for statistical analysis and data representation. Differences in TERT mRNA expression, TC, and telomerase activity between subgroups used the Wilcoxon rank-sum test. Associations of phenotype with clinical risk factors used Fisher exact test. Survival comparisons were by the Kaplan–Meier log-rank method using the R “survival” package (http://cran.r-project.org/web/packages/survival/index.html). Event-free survival (EFS) was calculated as time from initial diagnosis to the first occurrence of either relapse, progressive disease, secondary malignancy, or death. Patients without events were censored at the time of last follow-up. OS was defined as time from initial diagnosis until time of death from any cause, alive patients censored at last contact. Multivariate Cox regression was used to estimate the prognostic impact of TERT mRNA expression and C-circle presence while adjusting for ALK alterations, RAS/P53 alterations, ploidy, MYCN amplification, and 11q chromosome arm loss for both EFS and OS. Age and stage of the patients were not included in the final model, as all the patients in survival analysis were stage 4 high-risk patients with age at diagnosis >18 months.

See Supplementary Materials and Methods for additional details.

To comprehensively assess TMMs and their clinical relevance in high-risk neuroblastoma, we integrated the C-circle assay (n = 145), TC assay (n = 145), RNA-seq (n = 160), whole-genome sequencing (WGS; n = 136), WES (n = 222), and clinical covariates (data availability and overlap is presented in Supplementary Fig. S1). All specimens were obtained at initial diagnosis with written informed consent at COG member institutions in accordance with recognized ethical guidelines (24). Most cases assessed in this study were patients with stage IV high-risk neuroblastoma over 18 months of age (n = 107), with 3 <18-month-old high-risk patients and 24 patients being classified as low or intermediate risk. All samples analyzed had >70% tumor content on pilot sections and high-quality nucleic acids.

To investigate the prevalence of ALT in neuroblastoma, we quantified C-circles by qPCR for 134 primary tumors [110 high risk (30 MYCN amplified), 12 intermediate risk, and 12 low-risk] selected solely based on DNA availability (Table 1; Supplementary Table S1). Of the 107 high-risk tumors (age > 18 months), 25 [23.4%; 95% confidence interval (CI), 15.3%–31.4%] were positive for the ALT biomarker C-circles (Fig. 1A; Supplementary Table S1). For 23 of 25 ALT tumors WGS and/or WES data were available; ATRX genomic alterations were observed in 13 of 23 ALT tumors (57%; 10 focal deletions, two missense, and one nonsense mutations; Supplementary Table S1); nine focal deletions and the three single-nucleotide variants (SNV) were reported previously (24), while one sample with an ATRX focal deletion (PATCFL) was missed in the previous WES study (Supplementary Fig. S2A; ref. 24). None of the non-ALT tumors (C-circle negative) with WGS and/or WES (n = 102) had ATRX genomic alterations. We assessed C-circles in 11 additional tumor samples previously reported to have ATRX genomic alterations including one sample (PASAJU) with a focal gain, and all of them were positive for C-circles (Supplementary Table S1; ref. 24).

Given that ATRX deletions commonly appear as intragenic in-frame fusions (IFF; refs. 15, 33), we next explored the effect of ATRX variants on transcript levels using String Tie de novo assembly and quantification from RNA-seq and 42 different ATRX transcripts were predicted; we classified those transcripts into coding, likely coding, noncoding, exonic loss, and IFF (Supplementary Fig. S2B). Notably, aberrant transcripts matching IFFs were confirmed as the most expressed isoform in seven of 10 ATRX structural variant (ATRX^SV+) samples (PATGLU, PAPKXS, PANZVU, PANLET, PARACS, PATCFL, PASTCN, and PANLET), while the three remaining (PARYNK, PAPKXS, and PASRFS) expressed ATRX isoforms at very low levels (Supplementary Fig. S2C).

To identify additional ATRX-altered cases that were missed or not overlapping with the WGS/WES dataset, we compared the total expression of IFF-ATRX isoforms against ATRX-coding isoforms (Supplementary Fig. S2C). Two additional cases (PAIFXV and PANBJH) showed overexpression of IFF-ATRX, and both cases were positive for C-circles. Considering the combined DNA and RNA variant detection approaches, 10 ALT cases had focal deletions in ATRX, two cases had higher expression of IFF-like transcripts, and three had SNVs (Fig. 1A; Supplementary Fig. S3; Supplementary Table S1). A total of 10 of 25 ALT tumors had no ATRX genomic alterations (Fig. 1A; Supplementary Table S1), which we validated as wild-type for ATRX using Sanger sequencing. We did not detect genomic alterations in the other known ALT-associated genes (DAXX, H3F3A, or SMARCAL1) in ATRX wild-type ALT tumors (14, 17, 18). EFS and OS were similar among ALT tumors based on ATRX status (log-rank test; EFS, P = 0.741; OS, P = 0.346; Supplementary Fig. S4A and S4B).

To further validate activation of ALT in neuroblastoma without ATRX genomic alterations, we screened 104 neuroblastoma patient-derived cell lines and 28 neuroblastoma PDXs for C-circles using qPCR (Supplementary Table S2); five cell lines and three PDXs were C-circle positive. C-circle positivity was further confirmed by slot blotting of C-circle assay products onto a nylon membrane, detected using a DIG-labeled telomere probe (Supplementary Fig. S5A). APBs were also observed in ALT cell lines and PDXs (Supplementary Fig. S5B; ref. 34). The ALT cell line SK-N-FI was previously shown to be wild-type for ATRX (26, 35), and we identified two ALT PDXs (COG-N-589x and COG-N-620x) that were ATRX wild-type (Supplementary Fig. S5C; ref. 36). These ATRX wild-type ALT cell lines and PDXs were confirmed to express ATRX protein (Supplementary Fig. S5D and S5E). In addition, none of the ATRX wild-type ALT models had mutations in DAXX, H3F3A, or SMARCAL1. Thus, ALT can occur in neuroblastoma with and without ATRX genomic alterations.

ALT tumors were significantly associated with later disease onset (Supplementary Fig. S6A). Somatic mutation burden was significantly increased with patient age (Supplementary Fig. S6B) and in ALT tumors (Supplementary Fig. S6C). Consistent with prior reports, ALT tumors were exclusively non-MYCN–amplified (non-MYCN–Amp; Fig. 1A; Supplementary Table S3; refs. 35, 37), and these cases uniformly showed significantly lower MYCN mRNA expression (Supplementary Fig. S6D). Among the ALT cell lines and PDXs, two ALT cell lines had MYCN amplification (copy-number assay by qPCR; 32 copies for COG-N-515 and 33 copies for COG-N-512), but MYCN mRNA expression in these MYCN-Amp ALT cell lines was lower than other non-ALT MYCN-Amp cell lines (Supplementary Fig. S6D). We did not find any associations between ALT and segmental chromosomal alterations, regardless of ATRX status (Fig. 1A). Frequency of ALK, RAS/MAPK, and TP53/CDKN2A genomic alterations were also similar among ALT and non-ALT tumors (Fig. 1A).

In summary, approximately 23% of high-risk neuroblastoma tumors are ALT⁺. ALT activation can occur with and without ATRX alterations in neuroblastoma. ALT activation was detected exclusively in non-MYCN–Amp tumors and in relatively older patients.

As it is known that telomerase can be activated by TERT genomic rearrangements in neuroblastoma (5, 11), we analyzed recurrent SVs affecting the TERT locus in 85 neuroblastoma samples that had overlapping WGS, C-circle, and TERT expression data (Fig. 1B). Of 59 high-risk tumors (age > 18 months), 12 (20.3%; 95% CI, 10.1–30.6%) harbored structural rearrangements near the TERT gene (Fig. 1A; Supplementary Table S1), which is consistent with previous reports (5, 11). In addition, as reported previously (29), none of the tumors with WGS had TERT promoter mutations.

TERT expression quantified by RNA-seq versus qRT-PCR was strongly correlated in 19 primary tumors (Supplementary Fig. S7A; Spearman Rho = 0.72; P = 7.83 × 10⁻⁴). We assessed whether TERT mRNA expression could be used as an alternative to telomerase activity by using neuroblastoma cell lines for which TERT expression (by both qRT-PCR and RNA-seq) and telomerase activity (qTRAP by PCR ELISA) were simultaneously quantified (Supplementary Fig. S7B). TERT expression by RNA-seq strongly correlated with expression by qRT-PCR (n = 23; Supplementary Fig. S7C; Spearman Rho = 0.92; P = 2.57 × 10⁻¹⁰) and with telomerase enzymatic activity (n = 22; Supplementary Fig. S7D; Spearman Rho = 0.95; P = 6.226 × 10⁻¹²). Thus, RNA-seq can reliably quantify TERT mRNA expression and is indicative of telomerase activity.

As expected, TERT structural variant (TERT^SV+) tumors had the highest TERT mRNA expression, followed by MYCN-Amp tumors (Fig. 1C). ALT tumors expressed the least TERT of the high-risk subsets (Fig. 1A and C). Similar to patient tumors, ALT neuroblastoma cell lines and PDXs had low TERT expression (Supplementary Table S2) and were negative for telomerase activity (Supplementary Fig. S8A and S8B). The majority of high-risk tumors without TERT^SV+, MYCN amplification, or ALT had lower TERT expression when compared with MYCN-Amp or TERT^SV+ tumors (Wilcoxon P = 7.6 × 10⁻⁴; Fig. 1C), however, some had TERT expression as high as TERT^SV+ tumors (Fig. 1C), suggesting the existence of unidentified mechanisms for activation of TERT in high-risk neuroblastoma. In our cohort, none of the 24 nonhigh-risk tumors had high TERT expression (Fig. 1C).

Surprisingly, some MYCN-Amp tumors had TERT expression as low as seen in ALT tumors (Fig. 1C; Supplementary Fig. S9A). These MYCN-Amp samples had low TERT expression despite having high MYCN mRNA expression (Supplementary Fig. S9A). Similar to primary tumors, four MYCN-Amp cell lines had low TERT expression and were negative for telomerase activity (Supplementary Table S2; Supplementary Fig. S8B). Knockdown of MYCN expression using MYCN short hairpin RNA (shRNA) in a MYCN-Amp neuroblastoma cell line (COG-N-452h) decreased expression of TERT compared with nontargeted shRNA controls (Supplementary Fig. S9B; P < 0.05). Forced overexpression of MYCN using a lentiviral vector was undertaken in four non-MYCN–Amp neuroblastoma cell lines: two with telomerase activity (CHLA-171 and Felix-h) and two without telomerase activity (LA-N-6 and CHLA-132; Supplementary Fig. S9C). Forced MYCN overexpression upregulated TERT expression in one non-MYCN–Amp telomerase-positive cell line (CHLA-171), but failed to enhance TERT expression in the telomerase-positive cell line with a TERT rearrangement (Felix-h), or in the two cell lines negative for telomerase activity (Supplementary Fig. S9D; P < 0.005). This suggests that MYCN can upregulate TERT expression via an intact and nonrepressed TERT promoter, but MYCN overexpression by itself is not sufficient to overcome the TERT repression in TERT nonexpressing cells.

In summary, TERT^SV+ tumors, and the majority of MYCN-Amp tumors had high TERT mRNA expression. Some TERT^wt and MYCN^wt tumors had TERT expression as high as tumors with TERT^SV+, indicating some other unidentified mechanisms for activation of TERT in neuroblastoma. Not all MYCN-Amp tumors had high TERT expression. Consistent with that observation, forced overexpression of MYCN in telomerase-negative cell lines did not induce TERT expression.

We analyzed the outcome of 107 stage IV high-risk patients with age at diagnosis >18 months for which both TERT expression and ALT status were simultaneously assessed (Supplementary Table S1). High TERT expression, defined as ≥the median TERT expression of the high-risk cohort (n = 107), was associated with poor OS (P < 0.001; Fig. 2A) and poor EFS (P = 0.034; Supplementary Fig. S10A). The 5-year OS was 28% for the TERT-high (TERT-H) group versus 62% for the TERT-low (TERT-L) group. However, patients with ALT⁺ tumors showed no difference from non-ALT patients in OS (P = 0.99; Fig. 2B) or EFS (P = 0.32; Supplementary Fig. S10B).

Because the OS curves based on ALT status intersect, we computed the Cox regression HRs using time interaction term, and we observed a substantial increase in HR after 5 years for the ALT group (HR at >5 years, 14.1; P = 0.024). In non-ALT tumors, OS of high-risk patients was TERT expression dependent, poor with higher TERT expression (Supplementary Fig. S11A and S11B). As ALT tumors are a subgroup of TERT-L tumors (all ALT tumors had TERT mRNA expression below the median; Supplementary Fig. S9A), we combined analysis of ALT with TERT expression, which allowed us to define three groups of patients (TERT-H, ALT, and TERT-L/non-ALT) that each showed distinct OS probabilities (log-rank test comparing three groups, P < 0.001; Fig. 2C). The 5-year OS for TERT-H patients was 28% versus 46% for ALT patients and 75% for TERT-L/non-ALT patients, while 10-year OS was 22% for TERT-H, 24% for ALT, and 75% for TERT-L/non-ALT (Table. 2). All patients with TERT-L/non-ALT tumors who died had RIN values > 6.5, ruling out low RNA integrity as a cause for low TERT expression by RNA-seq. Consistent with previous studies, the presence of ALK and RAS/TP53 pathway alterations (Supplementary Fig. S12A and S12B; refs. 27, 38) and diploid DNA content (Supplementary Fig. S12C and S12D) were associated with poor outcome in these high-risk patients.

Unlike OS, EFS was not substantially different among the three groups defined by TERT expression and ALT status (P = 0.137; Fig. 2D). Therefore, we assessed OS in only those patients who experienced an event, and observed that while TERT-H and ALT patients had abysmal OS following an event, patients in the TERT-L/non-ALT group had significantly higher OS than patients in the TERT-H or ALT groups (log-rank test comparing three groups, P < 0.001; Fig. 2E). These data suggest that the differences in OS likely represent differences in the success of salvage therapy following an event.

Consistent with previous findings for high-risk patients, there was no difference in EFS (P = 0.68; Supplementary Fig. S13A) or OS (P = 0.46; Supplementary Fig. S13B) between MYCN-Amp and non-MYCN–Amp patients (39, 40). Even though the majority of MYCN-Amp tumors had high TERT expression and belonged to the TERT-H group (n = 22), some MYCN-Amp tumors had low TERT expression (n = 7; Supplementary Fig. S9A) and were classified as TERT-L. Notably, MYCN-Amp patients in the TERT-H group had a lower 5-year OS (29%) relative to the 86% seen for patients with MYCN-Amp and TERT-L tumors (P = 0.013; Fig. 2F). This observation was further validated in the SEQC dataset (https://hgserver1.amc.nl/cgi-bin/r2/main.cgi); MYCN-Amp tumors with low TERT expression had a significantly higher OS compared with MYCN-Amp tumors with high TERT expression (P = 0.0053; Supplementary Fig. S13C). Multivariate Cox regression analysis demonstrated that high TERT expression or ALT positivity predict poor OS but not EFS, independent of MYCN amplification, ploidy, ALK status, RAS/TP53 pathway alterations, or chromosome arm 11q deletion (Fig. 3A and B).

As median cutoff may not be optimal for distinguishing TERT-expressing and nonexpressing tumors for biological classification, we also defined a TERT expression threshold that separates TERT-positive and TERT-negative tumors by fitting a bimodal normal distribution (Supplementary Fig. S14A and S14B), as described previously (27). On the basis of the latter TERT expression threshold, 16 of 107 (14.9%) high-risk tumors were classified as TERT expression low and negative for the C-circle ALT marker (TERT&ALT⁻). OS but not EFS was significantly different for ALT, TERT-positive, and TERT&ALT⁻ tumors (log-rank test comparing three groups, P = 0.022; Supplementary Fig. S14C and S14D), results comparable with using median TERT expression for the cutoff (Fig. 2C).

Taken together, we classified patients with high-risk neuroblastoma into three subgroups based on TERT expression and ALT status. Non-ALT patients with lower TERT expression were associated with superior OS when compared with patients with ALT phenotype or higher TERT expression, irrespective of other known risk factors.

To investigate whether ALT is associated with high TC as described previously (14, 15, 26, 41), we assessed relative TC using telomeric qPCR in 134 neuroblastoma samples for which both ALT status and TERT expression were known. As expected, ALT tumors had higher relative TC than TERT-H (Wilcoxon P = 1.87 × 10⁻¹¹), TERT-L/non-ALT (Wilcoxon P = 9.5 × 10⁻⁴), or nonhigh-risk tumors (Wilcoxon P = 5.54 × 10⁻⁸; Fig. 4A). Consistent with primary patient tumors, ALT cell lines and PDXs had higher relative TC than non-ALT models (Supplementary Fig. S15A–S15C; Supplementary Table S2; Wilcoxon P = 2.55 × 10⁻⁵). ALT neuroblastoma cell lines were also associated with high and heterogeneous telomere length relative to telomerase-positive cell lines, as detected by TRF length analysis (mean TRF length >20 kb; Supplementary Fig. S15C) and by telomere FISH on metaphase spreads (Supplementary Fig. S15D).

TERT-L/non-ALT tumors and nonhigh-risk tumors were each significantly associated with a much higher relative TC when compared with TERT-H tumors (Wilcoxon P = 2.6 × 10⁻⁰⁷; Supplementary Fig. 4A). As WGS data were available for 85 of 134 samples, we measured telomere DNA abundance (TTAGGG/CCCTAA canonical repeats) from short-read WGS for each tumor and from a blood sample from each patient, and demonstrated a significant correlation between tumor TC measured by qPCR and WGS (Supplementary Fig. S16A; Spearman Rho = 0.701; P = 5.87 × 10⁻¹⁴). We observed that 11 of 17 TERT-L/non-ALT tumors had increased TC (tumor/blood ratio > 1; Supplementary Fig. S16B) and of 82 non-ALT high-risk tumor samples with age > 18 months, 16 tumors had high TC, similar in range to that observed within ALT tumors (relative TC > 5; corresponds to mean TRF length > 20 kb; Supplementary Table S1). This non-ALT high TC group meets the phenotypic criteria of the previously defined EST phenotype (26).

In the 16 tumors with high TC, 12 had low TERT expression and four had high TERT expression (Supplementary Table S1). To validate that TERT-L&C-circle⁻ tumors with high TC (TC > 5) are ALT negative, we assessed APBs in 34 primary patient tumors (12 tumors in the TERT-L&C-circle⁻ group with TC > 5, 11 tumors in TERT-L&C-circle⁻ group with TC < 5, and 10 C-circle⁺ tumors). APBs were detected in three of 12 TERT-L&C-circle⁻ tumors with TC > 5, while zero of 11 tumors in TERT-L&C-circle⁻ group with TC < 5 were APB positive. In C-circle–positive (C-cicrle⁺) tumors, eight of 10 were positive for APBs (Supplementary Table S1, Supplementary Fig. S17). Thus, the C-circle assay missed detection of ALT in three of 11 low TERT tumors with high TC, while the APB assay did not detect ALT in two of 10 C-circle⁺ tumors.

We compared clinical outcome between the 28 of 107 ALT tumors with ALT defined as either C-circle⁺ or APB-positive and the nine of 79 non-ALT tumors that had low TERT expression and TC > 5 (Fig. 4B). OS among these two groups was significantly different, with the ALT cohort having a much worse outcome (P = 0.007; Fig. 4C).

We also measured relative TC content, C-circles, and TERT mRNA in 104 neuroblastoma cell lines and 28 PDXs (Supplementary Table S2). We identified five neuroblastoma cell lines (LA-N-6, COG-N-291, CHLA-132, COG-N-509hnb, and COG-N-562hnb) and two PDXs (COG-N-518x and COG-N-649x) with low TERT expression that were ALT negative (Fig. 4D) and had relative TC > 5 (Supplementary Table S2), that is, consistent with the EST phenotype (26). We validated the lack of ALT activation in these models by confirming a low frequency of APBs (Supplementary Fig. S18A and S18B). All the EST cell lines and PDXs lacked telomerase activity by TRAP assay (Fig. 4E) and had high telomere length using TRF analysis in cell lines (mean TRF length >20 kb; Fig. 4F), and telomere/centromere FISH in PDXs (Fig. 4G) and showed high and heterogeneous telomere lengths (Fig. 4G; Supplementary Fig. S18C) with a lower frequency of telomere signal–free chromosomal ends than ALT cell lines (Wilcoxon P = 3.35 × 10⁻⁰⁷; Supplementary Fig. S18C and S18D). Unlike ALT tumors, where ATRX/DAXX genomic alterations are common (14, 15), none of the TERT-L/non-ALT tumor samples with TC > 5 (n = 14) had genomic alterations in ATRX, DAXX, H3F3A, or SMARCAL1. Consistent with the patient samples, the TERT-L/non-ALT cell lines and PDXs with TC > 5 expressed full-length wild-type ATRX and DAXX protein (Supplementary Fig. S19A and S19B).

We were able to induce the ALT-specific marker C-circles in the EST cell line LA-N-6 [but not in the telomerase-positive cell line SK-N-BE(2)] by stable ATRX knockdown using shRNA (Supplementary Fig. S20A and S20B). Knockdown of ATRX in LA-N-6 also increased APBs in these cells (Supplementary Fig. S20C and S20D). This suggests that ALT repressors such as ATRX can block activation of ALT in TERT-L/non-ALT tumors.

Thus, some neuroblastomas stratified clinically as high-risk have high TC despite no apparent activation of classical ALT markers or TERT expression, and as shown in Fig. 3C, this phenotypic subset has a much more favorable OS probability than the ALT tumors.

Roughly 15% of the high-risk neuroblastoma tumor samples were negative for TERT expression without any evidence of ALT activation (TERT⁻ and ALT⁻; Supplementary Fig. S14A). To investigate whether this phenotype was associated with yet another uncharacterized TMM, we measured the change in telomere length across multiple passages in three neuroblastoma TERT⁻ and ALT⁻ cell lines established from high-risk patients (LA-N-6, COG-N-346h, and COG-N-387h). All three cell lines had very low TERT expression, were negative for telomerase activity, and lacked ALT-specific markers (Supplementary Table S2); LA-N-6 is a non-MYCN–Amp neuroblastoma cell line with TC > 5, COG-N-346h is a non-MYCN–Amp neuroblastoma cell line with TC < 5, and COG-N-387h is a MYCN-Amp neuroblastoma cell line with TC < 5 (Supplementary Table S2). All three cell lines showed continuous shortening of telomere length by TRF analysis across progressive passages (Fig. 5A); shortening of telomeres was verified as a reduction in TC by qPCR (Fig. 5B). All three TERT⁻ and ALT- lines remained C-circle negative (Supplementary Fig. S21A) and TERT low across those same passages in culture (Supplementary Fig. S21B).

We observed a substantial decline in cell viability in two of three EST cell lines following several passages in vitro (Fig. 5C). Interestingly, the decline in viability of LA-N-6, a cell line with very high TC, was minimal even after 40 passages (Fig. 5C).

Taken together our data suggest that patients with TERT-L/non-ALT tumors derive a survival advantage from the failure of their tumors to maintain telomere length, but that some of these tumors may develop as yet unknown mechanisms to overcome this limitation.

We report here a comprehensive analysis of TMM in high-risk neuroblastoma in the context of tumor genomics and clinical outcome. Consistent with previous studies, we confirmed using the ALT-specific C-circle assay that ALT is activated in 23% of high-risk neuroblastoma tumors (26). ALT patient tumors, cell lines, and PDXs expressed very low TERT mRNA, indicating ALT and telomerase activation occur in a mutually exclusive manner.

ATRX genomic alterations have been associated with the ALT phenotype in neuroblastoma (15, 27), but the relationship of ALT (using a specific marker) to ATRX genomic alterations has not been precisely defined. We observed ATRX^SV+, IFFs, and SNVs in 60% of ALT tumors, but the remainder lacked any identifiable genomic lesions previously reported to be associated with the ALT phenotype. Consistent with previous reports that ATRX mutations are often enriched in older patients (15), the ALT patients in our cohort had a late disease onset irrespective of ATRX status. Patients with ALT phenotype had a poor OS, irrespective of ATRX status.

High TC has been associated with ALT (15, 37, 41), but we confirmed a previous report that long telomeres are not exclusive to ALT tumors in high-risk neuroblastoma (26), as 16% of the ALT-negative high-risk tumors had total TC as high as that seen in the ALT-positive tumors. Thus, assessment of ALT status using high TC as measured by FISH, WGS/WES, or qPCR would overestimate the frequency of ALT in neuroblastoma.

Similar to previous reports, we identified TERT^SV+ in approximately 20% of patients with high-risk neuroblastoma, and all TERT^SV+ tumors had highly upregulated TERT mRNA (5, 11). Among tumors with MYCN genomic amplification (MYCN-Amp), a subpopulation had TERT expression as low as that seen in ALT tumors. Using neuroblastoma cell lines, we showed that MYCN overexpression enhanced TERT mRNA expression in a TERT-expressing non-MYCN--Amp neuroblastoma cell line without TERT^SV+, but failed to induce TERT activation in cell lines with TERT^SV+ or in TERT nonexpressing cell lines. Thus, MYCN overexpression by itself might not be always sufficient to transcriptionally activate TERT.

As seen in a recent study (27), we observed that TERT activation in high-risk neuroblastoma was not limited to tumors with TERT^SV+ and MYCN amplification, as some tumors without these alterations expressed TERT as highly as TERT^SV+ and MYCN-Amp tumors. The prior study quantified TERT mRNA expression by microarray and detected ALT activation by scoring APBs on tumor tissue sections (27, 42), and reported that only three of 90 high-risk tumors lacked a TMM (42). By employing RNA-seq (validated by qRT-PCR) for TERT expression and the C-circle assay to detect ALT, we found that 12%–26% (depending on the cutoff employed) of high-risk neuroblastoma tumors (including some MYCN-Amp tumors) had low TERT expression and lacked ALT activation, and those patients had a significantly better OS. Consistent with this latter observation, high-risk neuroblastoma cell lines that were non-ALT, low TERT expressing, and lacked telomerase activity showed continuous shortening of telomeres with growth in vitro and two of three TERT-L/non-ALT cell lines spontaneously lost viability after multiple passages in culture.

By combining analysis of ALT status (by C-circles) and TERT mRNA expression as measured by RNA-seq we were able to classify patients with high-risk neuroblastoma into three subgroups based on TMM (TERT-H, ALT, and TERT-L/non-ALT). The 10-year OS in high-risk patients with TERT-H and ALT tumors was <25%, while the TERT-L/non-ALT group had a 75% 10-year OS, despite being classified as high-risk based on current risk stratification protocols (43). We also showed that OS for patients with MYCN amplification was significantly higher if the tumors had low TERT expression, which might explain the extreme dichotomy in the clinical course of MYCN-Amp patients (44).

We observed that TERT-high expressing and ALT neuroblastomas have poor clinical outcomes comprising uniform cohorts of true high-risk patients. An important and novel observation was our demonstration that 12%–26% of clinically high-risk neuroblastomas lacked a TMM and while those had EFS comparable with TERT-high and ALT patients, TMM-negative patients had a significantly higher long-term OS (75%). On the basis of our data, we hypothesize that TMM-negative tumors develop chemo-resistance and progress similarly to TMM-positive tumors, but due to a lack of TMM they continue to erode telomeres, which eventually enables salvage therapy given after disease progression to eradicate the tumors. Thus, current standard of care for patients with high-risk neuroblastoma that relies on dose-intensive cytotoxic chemotherapy in as short a time frame as possible may not be optimal for TMM-negative high-risk patients, and alternative strategies such as longer time frame low-dose chemotherapy should be tested in future clinical trials.

Employing the C-circle assay, considered to be specific for ALT cells and known to be a quantitative marker for ALT activity (13, 45), enabled us to identify ALT in 23% of patients with high-risk (age > 18 months) neuroblastoma, and these patients had a low OS (Fig. 2). However, three tumors with high TC were positive for APBs but had a C-circle content below cutoff, which suggested that some ALT tumors may be missed by implementing only the C-circle assay as marker for detection of ALT. We analyzed clinical outcome for patients with low TERT expression and high TC with ALT defined as C-circle positive or APB positive and we observed an apparent higher OS for TMM-negative patients identified by employing both assays to define ALT positivity (comparing Fig. 2C with Fig. 4C). APB analysis alone did not detect all ALT tumors as we identified two ATRX-mutant C-circle⁺ tumors that were APB negative, indicating that APBs have limitations in detecting ALT tumors, which is consistent with a previous report (14). In addition, APBs have the limitation of also being detectable in non-ALT cells (45). Cell lines with long telomeres resulting from forced telomerase overexpression have APBs (28), suggesting that APBs are associated with long telomeres rather than being specific for the ALT mechanism (45). Future studies in a large cohort of patients will be necessary to determine the optimal approach for TMM classification of tumors that have low TERT expression and a high TC.

Using the C-circle assay, we observed that patients with ALT-positive tumors had a dismal OS, with no difference in OS between ALT-positive and TERT-positive patients, which is consistent with previous reports that ATRX-altered high-risk tumors (all of which were ALT positive) were associated with very poor OS (15, 33). Roderwieser and colleagues reported a substantially better OS for ALT neuroblastoma compared with TERT-activated patients (42). We suggest that the difference in clinical outcome between our study and the prior study could be due to our use of the C-circle assay compared with use of APB's to identify ALT, differences in risk and age groups (our study focused on HR samples in patients with age > 18 months), and length of clinical follow-up, as ALT tumors tend to have an indolent phenotype (15, 46). In addition, our study here reported a higher frequency of ALT⁻and TERT⁻ tumors, which may be due to use of different methods to assess TERT, different TERT expression cutoffs, inclusion of only high-risk patients with age > 18 months, and differences in classification of low-TERT–expressing MYCN-Amp tumors.

A strength of our study is the relatively large sample size of patients with stage 4 high-risk neuroblastoma diagnosed at age > 18 months, representing patients receiving intense chemotherapy at high risk for a poor clinical outcome (43). Another strength of our study is that we determined TERT mRNA expression in the patient samples using RNA-seq, which may be more robust than microarray technology for quantifying low-abundance transcripts such as that of TERT, as suggested previously (47). A limitation of our study is that TERT expression is a continuous variable and cutoffs for high versus low expression using different approaches give comparable results. Our results need to be validated in a large cohort of patients, which will also allow optimizing the cutoff for TERT expression. While TERT mRNA expression could be affected by RNA integrity or tumor cell content, both of which were controlled for in this study, which is important to prevent classifying a tumor as TERT low inappropriately. Another limitation of our study and of previous studies that assessed telomere maintenance is that the patients studied were not from a uniformly treated group of patients receiving the most current standard-of-care therapies. Thus, it will be important to confirm our observations in a validation cohort of patients with high-risk neuroblastoma receiving uniform therapy on current protocols.

In summary, we demonstrated that quantifying TERT mRNA and C-circle abundance enables classifying patients with high-risk neuroblastoma into three subgroups that have distinct clinical outcomes, independent of currently employed risk stratification methods, including MYCN amplification. TMM-positive (TERT high expressing and ALT) neuroblastomas are true high-risk tumors, each with distinct, potentially targetable molecular pathways required for maintaining telomeres. Therapeutic strategies targeting telomerase and ALT activation warrant investigation in TMM-positive neuroblastomas (30, 48, 49). An important and novel observation is our demonstration that clinically high-risk patients with TMM-negative tumors have an exceptionally high OS, which is likely due to the eventual success of therapy (combined with telomere erosion) even after disease progression. If future studies confirm the latter observation, stratification of patients based on TMM using robust markers (TERT mRNA expression and DNA C-circles) will improve the analysis of future clinical trials, and a reduction in the intensity of therapy may be attainable for TMM-negative patients.

No potential conflicts of interest were disclosed.

Conception and design: B. Koneru, A. Farooqi, J.M. Maris, C.P. Reynolds

Development of methodology: B. Koneru, K.L. Conkrite, T.H. Nguyen, S.J. Diskin, C.P. Reynolds

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Koneru, A. Farooqi, S.J. Macha, J.L. Rokita, E. Urias, A. Hindle, H. Davidson, K. Mccoy, V. Yazdani, M.S. Irwin, D.A. Wheeler, C.P. Reynolds

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B. Koneru, G. Lopez, T.H. Nguyen, A. Modi, J.L. Rokita, S. Yang, J.M. Maris, S.J. Diskin, C.P. Reynolds

Writing, review, and/or revision of the manuscript: B. Koneru, G. Lopez, A. Farooqi, K.L. Conkrite, T.H. Nguyen, J.L. Rokita, A. Hindle, M.S. Irwin, J.M. Maris, S.J. Diskin, C.P. Reynolds

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Davidson, K. Mccoy, J. Nance, M.S. Irwin, C.P. Reynolds

Study supervision: J.M. Maris, S.J. Diskin, C.P. Reynolds

We thank the NCI Pediatric Preclinical Testing Consortium for providing some of the PDX sequencing data, the Children's Oncology Group (COG) Neuroblastoma Biobank for tumor samples and clinical data, and the COG Childhood Cancer Repository (www.CCCells.org), supported by Alex's Lemonade Stand, for providing cell lines and PDXs for this study. We thank the patients and their families for donating samples to enable this research. This work was supported by grants from Cancer Prevention & Research Institute of Texas RP170510 (to C.P. Reynolds), NCI CA217251 (to C.P. Reynolds), NCI CA221957 (to C.P. Reynolds), NCI R01CA204974 (to S.J. Diskin), NCI R35CA220500 (to J.M. Maris), the U.S. NIH grants RC1MD004418 to the TARGET consortium, CA98543 and CA98413 to the Children's Oncology Group, Alex's Lemonade Stand Foundation for support of the COG Childhood Cancer Repository (www.CCcells.org), and the Roberts Collaborative for Genetics and Individualized Medicine (to G. Lopez).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.