Pan-Cancer Analysis of Clinical Relevance via Telomere Maintenance Mechanism

Understanding the telomere maintenance mechanism (TMM) in immortal cancer cells is vital for TMM-targeted therapies in clinical settings. In this study, we classified four telomere maintenance mechanisms into telomerase, ALT, telomerase + ALT, and non-defined telomere maintenance mechanism (NDTMM) across 31 cancer types using 10,704 transcriptomic datasets from The Cancer Genome Atlas. Our results demonstrated that approximately 50% of the total cohort displayed ALT activity with high telomerase activity in most cancer types. We confirmed significant patient prognoses according to distinct TMMs in six cancer types: adrenocortical carcinoma (ACC), PAAD, HNSC, SARC, GBM, and metastatic cancer. Patients with metastasis had a poor prognosis in the ALT group (p < 0.006) subjected to RAS protein signal transduction. Glioblastoma patients had poor prognosis in NDTMM (p < 0.0043) and showed high levels of myeloid leukocyte activation. Pancreatic adenocarcinoma (p < 0.04) and head and neck squamous cell carcinoma (p < 0.046) patients had a good prognosis in the ALT group with high immune cell activation. Furthermore, we showed that master transcriptional regulators might affect the selection of the TMM pathway and explained why different telomere maintenance mechanisms exist. Furthermore, they can be used to segregate patients and predict responders to different TMM-targeted therapeutics.


Introduction
The telomere maintenance (TMM) mechanism is used by cancer cells to promote immortality [1]. Recently, as the research on telomerase [2] and alternative lengthening of telomeres [3] in human cancer is being actively conducted, interest in the role of TMM in the immortality of tumor cells (which is one of the hallmarks of cancer) is increasing. It has also been studied in cancer cell lines with tumors of relevant origin based on TERT isoform expression patterns [4].
In most cancer cells, telomerase activity is maintained; however, some cancer cells, such as telomerase-deficient cancer cells, use an alternative lengthening of the telomeres (ALT) mechanism for their survival [5]. Telomere lengthening is mainly mediated by TERT (telomerase) and ALT (ATRX/DAXX alteration); however, in approximately 22% of the samples, a non-defined telomere maintenance mechanism (NDTMM) might be involved [6]. Little is known about the NDTMM, but it has been reported in several cancer types, including glioblastoma [7], osteosarcoma [8], and metastases of cutaneous melanoma [9]. In addition, the role of telomere homeostasis in metastatic cancer is unknown, and targeting TMM in aggressive metastatic tumors with a poor prognosis can be a good strategy. 2 of 11 Therefore, it is crucial to understand the molecular mechanisms underlying the four types of telomere maintenance mechanisms and their impact on the survival of patients.
To study the clinical relevance of the four TMM-associated pathways, we performed a thorough assessment of their relationship with clinical prognostic indicators in various cancer types. We have comprehensively analyzed ALT activities across 31 cancer types in The Cancer Genome Atlas (TCGA) [10]. In this study, we primarily focused on the distinct molecular features related to the four TMM types and assessed their clinical relevance. The following results were obtained: First, cancer types differ significantly in prognosis according to the four TMM types. Second, various cancer types have different molecular profiles depending on the type of TMM; TMM types in pan-cancer are associated with genomic alternations [11]. Third, certain TMMs are only associated with different biological processes. The functional diversity of telomerase indicates important differences between these two TMM pathways (telomerase and ALT), which may prove to be essential in cancer for the acquisition of metastatic phenotypes [12].
Therefore, our goals were to refine our understanding of the four TMM types and use this framework to identify drug targets that can be harnessed to overcome TMM-type-based resistance.

Telomere Maintenance Mechanism Separated Patient Outcome
To classify the telomere maintenance mechanisms, we used TCGA RNA-sequencing data from 31 cancer types with pooled metastatic tumor samples from 11 cancer types ( Figure 1A, Supplementary Data Table S1). Four telomere maintenance mechanisms were defined [10] according to the TMM signature [13]. To classify the TMM subtype using transcriptome data, we used a single-sample gene enrichment score for a single patient sample (ssGSVA) [14]. Then, we split the samples into four types: telomerase, telomerase + ALT, ALT, and NDTMM samples for each cancer type ( Figure 1A). Among 10,704 samples, 47% displayed both telomerase and ALT, 27% displayed ALT, 9% telomerase, and 17% NDTMM [6] ( Figure 1B). The four TMM activities varied across the cancer types. Cholangiocarcinoma (CHOL) showed no telomerase activity among the 31 cancer types. We also calculated the telomere maintenance mechanism in metastatic tumor samples from the TCGA, with 11 cancer types. The frequency of telomere maintenance mechanism types in metastatic tumors was similar to that of primary tumor samples ( Figure 2C). In five cancer types, namely, adrenocortical carcinoma (ACC), pancreatic adenocarcinoma (PAAD), head and neck squamous cell carcinoma (HNSC), sarcoma (SARC), and glioblastoma multiforme (GBM), the four types of telomere maintenance mechanism presented significant prognostic value (SARC: p = 7.4 × 10 −3 , ACC: p = 4.0 × 10 −2 , GBM: p = 4.5 × 10 −2 , PAAD: p = 4.0 × 10 −2 , HNSC: p = 4.6× 10 −2 ) ( Figure 1D). GBM had a poor survival rate for NDTMM [15], and the ALT groups of GBM showed poor survival rate. Although NDTMM has only been reported in certain cancer types [1], our results showed that NDTMM could function in all cancer types.
In contrast, ACC with NDTMM had a good survival rate. ACC and GBM with ALT displayed the opposite trend regarding survival rate. Two cancer types with the telomerase mechanism, PAAD and HNSC, had poor outcomes. In addition, high ALT levels (p = 0.04) were associated with a better prognosis of PAAD. Overall, our analyses showed that TMM type might distinguish patient prognosis in a single patient sample and can be used as a prognostic marker.   Next, analysis of transcriptional factor (TFs), as master regulators, indicated that the favorable risk with NDTMM may be regulated by the transcriptional factors NFKB1, RUNX3, SPI1, and POLR2A (FDR = 0.0001) ( Figure 2E), whereas the unfavorable risk with ALT may be regulated by SETDB1, CBX3, HCFC1, TCF7L2, and STAT1 (FDR = 0.0001) in the "Ras protein signal transduction" pathway, including the gene RIT1 in "RET signaling" pathway ( Figure 2F). Expression of E2F1 ( Figure 2G), as an hTERT repressor TF, was significantly different between ALT (p = 6.4 × 10 −5 ) and NDTMM. We then assessed the frequency of pathologic tumor stage in the four distinct TMM groups. Interestingly, NDTMM was highly associated with grade 3, but NDTMM was present in a small fraction of grade 4 ( Figure 2H). These results demonstrate that TMMs may differentially contribute to tumor progression of metastatic cancer. Overall, TMM types in metastatic cancer have a frequency similar to that in primary tumors, but master regulators and signaling pathways in ALT were different from those reported in a previous study. Therefore, the type of TMM may be a useful prognostic marker in patients with metastatic cancer.

Molecular Characteristics Based on the Four TMM Types
ATRX and DAXX gene mutations might be more generally associated with the ALT phenotype [18], and TERT promoter mutations enhance telomerase activation [19]. We focused on six cancer types with distinct prognoses and identified their molecular characteristics in each specific TMM type.
Repair of dysfunctional telomeres by fusion propels cells into breakage-fusion-bridge cycles, resulting in unequal distribution of genetic material into daughter cells, and, hence, genome instability [20]. Telomere dysfunction increases mutation rates and genomic instability [21]. Next, we analyzed the copy number variation and tumor mutation burden profiles of 1201 primary cancer specimens and 395 metastatic cancer specimens across six cancer types with pooled metastatic tumor samples.
ACC and SARC showed significantly higher copy number variations in the ALT group than in the NDTMM group ( Figure 3A). Interestingly, several cancer types, including HNSC and PAAD, displayed a high tumor mutation burden (TMB) in the four TMM types. Metastasis cancer showed a similar pattern to ACC in four TMM types. For ACC and metastatic cancer, the patient prognosis according to TMMs was the same. We found that high copy number variation (CNV) in ALT was associated with poor prognosis for ACC, SARC, and metastatic cancer ( Figure 2B). The five cancer types showed significantly different mutation frequencies between ALT and NDTMM. KRAS was the most frequently mutated gene in ALT and TP53 was frequently mutated in NDTMM for five cancer types ( Figure 3B). We confirmed a significant difference in stemness (p < 0.0007) between ALT and NDTMM in metastatic cancer. Both telomerase and ALT activity may cause high stemness in GBM ( Figure 3C). This result suggests that in six types of cancers a specific telomere maintenance mechanism is associated with genomic instability of copy number variation and mutation during cellular proliferation. In particular, in the case of ALT, it was confirmed that the prognosis was poor compared to other TMM types; ALT type was associated with relatively high copy number variation, and this result may provide a critical clue to the synthesis of non-canonical telomeric DNA. Together, these studies indicate that subtelomeres are hotspots of DNA breakage and repair, and are likely to be responsible for the generation of complex interchromosomal duplication patterns and the rapid evolution of these genomic regions, as well as the prevalence of large CNVs near telomeres [22].

Different Biological Processes Affected Patient Prognoses of Different TMM Groups
We confirmed different patient prognoses according to the four TMM types ( Figure  1D). We performed a gene ontology analysis to obtain functional insights into TMM types according to clinical outcomes. Poor outcome-related biological pathways enriched in ACC with ALT were related to peptide secretion, purine-containing compound metabolic process, mitochondrion organization, and interferon-gamma production ( Figure 4A).
Next, we analyzed the transcription factors and target gene networks. The mitochondrion organization, mitochondrial respiratory chain complex assembly, organelle disassembly, and regulation of ketone biosynthetic processes (FDR = 0.001) were enriched in ACC with ALT ( Figure 4B). Mitochondrial biogenesis was higher in the ALT group than in the telomerase group according to a previous study [23]. The vulnerability of the mitochondrial genome to mutations and the somatic mutations promote poor prognosis [24].

Different Biological Processes Affected Patient Prognoses of Different TMM Groups
We confirmed different patient prognoses according to the four TMM types ( Figure 1D). We performed a gene ontology analysis to obtain functional insights into TMM types according to clinical outcomes. Poor outcome-related biological pathways enriched in ACC with ALT were related to peptide secretion, purine-containing compound metabolic process, mitochondrion organization, and interferon-gamma production ( Figure 4A). and FOXM1 genes in HNSC with ALT ( Figure 4G). In GBM, the poor outcome related to NDTMM and was enriched in myeloid leukocyte activation, TNF signaling pathway, PDGFRB pathway, ROS, and RNS production in phagocytes ( Figure 4H). Favorable risk for GBM is related ALT and enriched in cell cycle, DNA replication ( Figure 4I).
Overall, our analyses showed that different biological processes might affect the four TMM types in an individual sample of a specific cancer type.  Next, we analyzed the transcription factors and target gene networks. The mitochondrion organization, mitochondrial respiratory chain complex assembly, organelle disassembly, and regulation of ketone biosynthetic processes (FDR = 0.001) were enriched in ACC with ALT ( Figure 4B). Mitochondrial biogenesis was higher in the ALT group than in the telomerase group according to a previous study [23]. The vulnerability of the mitochondrial genome to mutations and the somatic mutations promote poor prognosis [24].
Overall, our analyses showed that different biological processes might affect the four TMM types in an individual sample of a specific cancer type.

Discussion
The telomere maintenance mechanisms play essential role in the immortalization of cancer cells, and tumor cell survival is mainly maintained by two mechanisms: telomerase and alternative lengthening of telomeres. In a previous study, 22% of all TCGA cancers did not express TERT or had mutations in ATRX or DAXX [6]. The frequency of ALT occurrence varies by cancer types. A higher rate of ALT activation was reported in tumors of mesenchymal origin than in carcinomas of epithelial origin. However, the reason for this is still not clearly known [25].
It has been reported that about 19% of ALT gastric cancers occur in tumors with MSI high, but it has been recently reported that about 30% of gastric cancer occur ALT in the stem-like molecular type [43], suggesting that ALT frequency may depend on the molecular subtypes. Pertinent to this, since the frequency of ALT activity may be different for each molecular subtype in individual cancer types, there would be discrepancy between observed and predicted ALT frequency according to the composition of subgroups in population of evaluation. Thus, it might explain a difference between the previously reported frequency of ALT activity and that predicted in this study.
In this study, we showed that NDTMM occurs in 30 cancer types (96.77%). In ACC, SARC, and metastatic cancer, samples with NDTMM had the best prognosis, but the prognosis was poor in the ALT group. We confirmed that ALT in metastatic cancer is related to the RAS protein signaling pathway.
This suggested that ALT could use different signaling pathways for each cancer type.
In the ALT groups of ACC, SARC, and metastatic cancer, poor outcome-related molecular profiles were associated with significantly higher CNV. PAAD and HNSC showed relatively good prognosis in the telomerase group, and high immune cell activation, such as antigen-presenting cells and myeloid leukocyte activation, was confirmed in the ALT group. This is the first study to show that two cancer types, PAAD and HNSC, have a better prognosis in the telomerase group than in the ALT group. In addition, we confirmed that higher TMM activation was associated with higher stemness in metastatic cancer. Alternative lengthening of telomeres is important for epidermal homeostasis and tumorigenesis in cancer stem cells [48]. Although our study is limited to bioinformatics analysis and has a limited number of samples depending on the type of TMM, future study is needed to assess associated candidate pathway genes for six cancers associated with TMM types.

Telomere Maintenance Mechanism Classification
To test the telomere maintenance mechanism, we used single-sample gene variation analysis (ssGSVA) [14] of 31 RNA-seq data from TCGA. The TCGA mRNA expression dataset was obtained from Broad GDAC Firehose (https://gdac.broadinstitute.org/, accessed on 1 August 2020). The gene set used to evaluate TMM was the same as that used in a previous study [13]. We performed 100,000 or more runs to increase the statistical significance. We classified four types of TMM per cancer, and the criteria for classification was to find the TMM with the highest relative activity among the four types and identify the sample. TEL, relatively high telomerase activity; ALT, ALT activity; NDTMM, non-defined telomere maintenance mechanism with no or low telomerase activity; and TEL+ALT, ALT activity with telomerase activity.

Differential Expression Gene Analysis in Cancer Types
We performed DEG analysis for the good outcome samples compared to the poor outcome samples, as well as the samples with NDTMM compared to the samples with ALT in six cancer types (ACC, GBM, HNSC, PAAD, SARC, and metastatic cancer) using the "Limma" R package [49].

Survival Probability Analysis and Gene Ontology and Correlation Analysis
The R package "survival" [50] was used to perform the overall survival analysis and produce the Kaplan-Meier survival plots. A log-rank test was used to assess the statistical significance (p < 0.05). Gene ontology analysis was performed using METASCAPE [51] and DEGs (FDR < 0.05).

Transcription Factor Analysis Protein Association Network
We identified transcription factors (TFs) and target genes using the Cytoscape plug-in iRegulon, which pairs motifs and chromatin immunoprecipitation-sequencing (ChIP-seq) tracks to determine the TFs controlling gene networks, and the iRegulon database (version 2015.02.12) [52]. We focused on six main TMM pathways and signature gene sets [13].