Mesenchymal and MAPK Expression Signatures Associate with Telomerase Promoter Mutations in Multiple Cancers

Stratification of cancer patient tumors according to genetic alterations in a tissue-agnostic manner is emerging as a valuable tool for directing therapies (1). Many cancers harbor heterozygous C>T transitions in the proximal promoter for telomerase reverse transcriptase (TERT; refs. 2–4). The human TERT gene encodes the catalytic subunit of telomerase (5), which maintains telomere length in stem cells and most cancer cells (6–9). TERT promoter (TERTp) mutations drive allele-specific expression (10, 11) and are especially prevalent in glioblastomas, melanomas, myxoid liposarcomas, liver cancers, thyroid cancers, and bladder cancers (4, 12, 13). For unknown reasons, TERTp mutations frequently associate with poorer patient survival (14–17).

Mutant TERT promoters display distinct regulatory features (18, 19), but the mechanistic details regulating transcription from mutant TERT promoters are incompletely understood. The mutation creates a de novo consensus-binding sequence for E-twenty-six (ETS) transcription factors, a large family with more than 26 members. In glioblastoma, liver cancer, and bladder cancer cell lines, the housekeeping ETS transcription factors GABPα & GABPβ1 are recruited to the mutant TERT sequence (11, 18). In glioblastoma, ETS1, and in thyroid cancer, ETV5, appear to play roles in these specific cancers (20, 21). These specific ETS factors have not been found at wild-type TERT promoters (11, 18), and published data suggest that such promoters generally may be reliant on a different subset of factors (13, 22, 23). Importantly, the specific isoform GABPβ1L was shown in mice with TERTp-mutant xenografts of glioblastoma to regulate survival in a TERT-dependent manner (24) demonstrating that these factors drive aggressiveness in this established model of brain cancer. In addition to ETS factors, initial studies indicate that some TERTp-mutant cancer types rely on RAS pathway signaling to maintain TERT mRNA expression (3, 21, 25–28), but the generality of this observation has not been established.

Our current study was motivated to understand (i) the high frequency of TERT promoter mutations in certain cancer types and their complete absence in other cancer types (4, 13, 22), and (ii) the poorer patient outcomes that often associate with these mutations. It has been observed that TERTp mutations are common in cancers that arise from more slowly proliferating cell types and may provide a proliferative advantage (29). To more fully explore the underlying biological associations with TERTp mutations across cancer types, we employed a pan-cancer analysis including hundreds of tumor-derived cell lines from 19 tissue types. We hypothesized that TERTp mutations may provide a selective advantage in, or be necessitated by, specific cellular programs. We considered that such programs, if functionally linked with TERTp mutations, could be revealed by analysis of gene and protein expression profiles across multiple cancer types. Our expectation from these studies was to determine whether common cellular mechanisms operate in TERTp mutants that distinguish them from cells that lack these mutations, both biologically and clinically.

Our approach incorporated three interrelated types of analysis: (i) using gene-set enrichment analysis to analyze the differences between TERTp mutant versus wild-type cell lines, (ii) generation of transcriptional signatures of TERTp mutations, including a TERT signature, and comparing these signatures versus an independent signature of TERT mRNA activation, and (iii) experimental validation of several findings focusing on specific epithelial-to-mesenchymal (EMT) gene markers.

The map of cellular states (Onco-GPS map) displayed in Fig. 2 and Supplementary Fig. S1 was obtained following the Onco-GPS methodology (30). The Onco-GPS is a PanCancer map very similar to the one shown in Supplementary Fig. S1 of our previous study (30). The samples on the map are color coded to indicate the mutation status of TERT (Fig. 2B) or the single-sample GSEA enrichment score of TERT transcriptional signatures (Fig. 2C and D; Supplementary Fig. S1B and S1C) or Hallmark gene sets (Fig. 2A; Supplementary Fig. S1A).

RNA sequencing (RNA-seq), Affymetrix, and reverse-phase protein array (RPPA) data were analyzed from the Cancer Cell Line Encyclopedia and TERT promoter mutation genotype calls were annotated as described previously (10, 31). For Fig. 3, 503 lines and for Supplementary Fig. S3, 278 lines were analyzed for which the TERT promoter mutation status and RNA-seq data were determined. For GABPβ1 isoform analysis, transcript level expression (January 2, 2019 version) were downloaded from the CCLE data portal (https://portals.broadinstitute.org/ccle/data). Comparisons were performed using a Wilcoxon rank sum test.

Data for cell line drug response (EC₅₀) were downloaded from the Cancer Therapeutics Response Portal (CTRP v2: ftp://caftpd.nci.nih.gov/pub/OCG-DCC/CTD2/Broad/CTRPv2.0_2015_ctd2_ExpandedDataset). Cell lines were stratified into TERT promoter mutant and wild-type. Cell lines with missing EC₅₀ values were excluded from comparisons on a drug-by-drug basis. Comparisons were performed using a Wilcoxon rank sum test.

SNU-423, SNU-475, SNU-398, HEK293T, and DAOY were obtained from the ATCC. U87MG, SCaBER, and MDA-MB-231 were obtained from the University of Colorado, Anschutz (Aurora, CO), Tissue Culture Shared Resource. HaCaT cells were a gift from X. Liu. Mel3249, Mel3616, and Mel1692 were gifts from K. Couts at the University of Colorado, Anschutz Medical Campus (Aurora, CO; refs. 32, 33). Cancer cell lines and HaCaT cells were cultured in DMEM (VWR Scientific) with 2 mmol/L GlutaminePlus (Atlanta Biologicals), 10% FBS (Thermo Fisher Scientific), 2 mmol/L GlutaMAX-I (Gibco), 100 U/mL penicillin and 100 mg/mL streptomycin (Gibco) and 1 mmol/L sodium pyruvate (Gibco). Cell lines were tested for Mycoplasma contamination by the supplier or by PCR (MD Biosciences) at the University of Colorado Tissue Culture Facility and were used for experiments within 10 weeks of resuscitation. Cell lines from ATCC and the University of Colorado Tissue Culture Resource were authenticated by short tandem repeat analysis by the manufacturer or at the Heflin Center for Genomic Sciences at University of Alabama (Birmingham, AL). HEK293T cells were grown to 60% confluency in a 10-cm plate and transfected with 6 μg of siSNAI2#3-shRNA-pLKO.1 plasmid (Addgene #10905; ref. 34), 3 μg of pDelta-8.9 plasmid containing gag, pol, and rev genes, and 0.6 μg of a plasmid encoding VsVg (Functional Genomics Facility, University of Colorado, Anschutz Medical Campus, Aurora, CO). Cells were transfected with Lipofectamine 2000 Transfection Reagent (Life Technologies/Invitrogen 11668-019). Media were changed after 12 hours, followed by incubation for 48 hours in 20 mL. Media were harvested and filtered with 0.2-μm filters and 8 mL was added to a 10-cm plate of cell lines for 24 hours, followed by fresh media for 48 hours at which point the cells were selected with puromycin until colonies emerged. Puromycin concentrations per mL used were 1 μg for DAOY, 1.2 μg for U87, 0.8 μg for MEL3616 and MEL3429, and 1.5 μg for MDA-MB-231.

Following RNA extraction with TRIzol (Life Technologies), reverse transcription was performed by treating 10 μg of RNA with 5 units of RQ1 DNase (Promega) according to the manufacturer's protocol, followed by phenol extraction (pH 6.7, Amresco #0883), then chloroform: isoamyl alcohol extraction (VWR #X205), and then 70% ethanol precipitation. The cDNA was then generated from 2 μg of RNA synthesized using random hexamers, oligo (dT) 20-mer, and SuperScript III (Life Technologies). Following treatment with RNase H (New England Biolabs), quantitative PCR was performed with either SybrSelect (Thermo Fisher Scientific) or iQ SYBR Green (Bio-Rad) PCR mix using a Roche LightCycler 480 with the program 10 minutes at 98°C, 30 seconds at 95°C, 30 seconds at 60°C, 30 seconds at 72°C, and 5 minutes at 72°C, followed by quantification using the Roche LightCycler 480 software. Melt curve analyses were examined to ensure the uniformity of relevant PCR amplicons and all PCR amplicons were sequenced at least once to confirm the product identity. Primers for TERT cDNA exon 2 were forward 5′-CGTGGTTTCTGTGTGGTGTC-3′, reverse 5′-CCTTGTCGCCTGAGGAGTAG-3′; and TERT cDNA exon 14 were those described previously (17). Primers for FOS cDNA were forward 5′-AGAATCCGAAGGGAAAGGAA-3′, reverse 5′-CTTCTCCTTCAGCAGGTTGG-3′; for GABPβ1L cDNA (24) were forward 5′-ATTGAAAACCGGGTGGAATC-3′, reverse 5′-CTGTAGGCCTCTGCTTCCTG-3′; for CDH1 cDNA were reverse 5′-GAACGCATTGCCACATACAC-3′, reverse 5′-ATTCGGGCTTGTTGTCATTC-3′; for GAPDH cDNA were forward 5′-CTGCACCACCAACTGCTTAG-3′, reverse 5′-GTCTTCTGGGTGGCAGTGAT-3′.

After removing media from cells in a 6-well plate, 1.5 mL of ice-cold PBS was added and the cells were scraped into a 1.7 mL tube. Cells were collected by centrifugation at 500 × g for 4 minutes, PBS was removed, and cells placed on ice for lysis, or snap-frozen in LN2 and stored at −80°C. Cell pellets were lysed in 10 mmol/L Tris-Cl (pH 8.0), 150 mmol/L sodium chloride, 1% Triton X-100, 1 mmol/L EDTA at 750,000 cells per 15 μL. Lysis buffer contained per 100 μL, 3 μL of Complete Protease Inhibitor (Sigma #P8340), and 3 μL benzonase (Sigma-Aldrich E1014). Samples were incubated on ice for 10 minutes. Samples were made up to 1× LDS loading buffer and 5% 2-mercaptoethanol (βME), incubated at 95°C for 7 minutes before loading 15 μL onto NuPAGE Novex 4–12% Bis-Tris polyacrylamide gels (Thermo Fisher Scientific, NP0321). Gels were run in 1× MES buffer for 35 minutes at 200 V. Gels were transferred to GE Healthcare Amersham Hybond -N+ Membranes (Thermo Fisher Scientific 45-000-927) for 1 hour at 4°C at 0.5 amps. Membranes were blocked in 5% nonfat dry milk in PBS-T (PBS + Tween 20, 0.05%) or Starting Block Blocking buffer (Thermo Fisher Scientific, catalog number 37539) with orbital shaking for 1 hour at room temperature or several hours at 4°C. Membranes were then cut into strips for appropriate sizes depending on the antibody. Antibodies were incubated with blots for 1 to 8 hours at 4°C with orbital shaking, followed by antibody removal and rinsing of the membranes twice with 5 mL PBS-T followed by washing in 10 mL of PBS-T three times for 15 minutes at room temperature with orbital shaking. Primary antibodies were detected by addition of species-specific horseradish-peroxidase (HRP) conjugated to secondary antibody in blocking buffer, incubated with orbital shaking for 30–60 minutes followed by washing as for primary antibodies. After removing the last wash buffer, chemiluminescent visualization solution was added. For chemiluminescent detection, SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, catalog number 34096) or SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Scientific, catalog number 34577) was used. Membranes were visualized with Alpha Imager.

Antibodies used for immunoblots were ERK1/2 [p44/42 MAPK (Erk1/2) antibody Cell Signaling Technology, catalog number 9102S], phospho-ERK1/2 [pERK; phospho-p44/42 MAPK (Erk1/2; Thr202/Tyr204; D13.14.4E XP (R) Rabbit mAb Cell Signaling Technology 4370S], and SNAI2 (C19G7 Rabbit mAb, Cell Signaling Technology 9585S). Dilutions for ERK, pERK, and SNAI2/SLUG were 1:1,000 for both primary and secondary antibodies. For housekeeping genes, dilutions for both primary and secondary antibodies for GAPDH (D16H11 XP rabbit mAb, Cell Signaling Technology 5174S) and histone 3 (Abcam ab1791) were 1:5,000.

Chromatin immunoprecipitation (ChIP) was performed as described previously (11). For immunoprecipitation, 10 μg of solubilized chromatin was used with 5 μL of α-SNAI2/SLUG antibody (C19G7 rabbit mAb, Cell Signaling Technology 9585S) or an equivalent mass of nonspecific IgG control (12-370, EMD Millipore), and nutated overnight at 4°C. The antibody for SNAI2/SLUG ChIP has been validated by a previous study (35). Protein G/Protein A agarose beads (IP05-1.5 mL, EMD Millipore Corporation) were added for three hours and then treated as described previously (11). PCR analysis was performed for the TERT promoter as described previously (11) with the forward primer 5′-GTCCTGCCCCTTCACCTT-3′ and reverse primer 5′-AGCGCTGCCTGAAACTCG-3′, and for TERT intron 1/exon 2 boundary using primers forward 5′-GCAGGTGTCCTGCCTGAA-3′ and reverse 5′-GAAGGCCAGCACGTTCTTC-3′, or exon 2 with the forward primer 5′- CTACTCCTCAGGCGACAAGG-3′ and reverse primer 5′- TGGAACCCAGAAAGATGGTC -3′.

We compared the expression profiles of a subset of cancer cell lines from the Cancer Cell Line Encyclopedia (31,36) harboring TERTp mutations (83 samples in 13 tissue types) against wild-type (419 samples in 16 tissue types) to identify differentially expressed genes. Then we used single-sample gene set enrichment analysis (ssGSEA; ref. 37) and gene sets from the Molecular Signatures Database (MSigDB.org; ref. 38) to identify gene expression patterns in the TERTp cell lines. For our analysis, we considered the set of distinct TERTp mutations shown in Supplementary Table S1.

The analysis strongly suggested that the specific transcriptional profile of TERTp-mutant cell lines is that of a mesenchymal/EMT cell type. This can be seen, for example, in Fig. 1A, which shows the top-scoring gene sets from the hallmark collection of the Molecular Signatures Database (MSigDB.org). Each hallmark in this collection of 52 gene sets consists of a "refined" gene set, derived from multiple "founder" sets, that conveys a specific biological process and displays coherent expression (38). The top hit is the hallmark for the epithelial-to-mesenchymal transition (EMT). We then matched a collection of gene sets representing major oncogenic pathway components that we previously introduced (30) and observed that the top two hits were transcriptional components C6, representing BRAF/MAPK activation, and C4 representing core EMT signaling (Fig. 1B).

We performed a second analysis using a broader range of gene sets from the C2 subcollection of MSigDB plus additional gene sets from the literature (5,334 gene sets). Among the top 30 hits are two independent gene sets representing EMT, one gene set representing BRAF and MITF activation signatures, one gene set representing the response to TGFβ1, and several representing invasive and metaplastic breast cancers, stemness, and metastasis (Fig. 1C). Interestingly, the top hit is gene set GDS337_hTERT derived from a TERT-rescue of late-passage mammary epithelial cells (HMEC; ref. 39).

To gain a more global and contextual perspective of the association between TERTp, mesenchymal characteristics, and BRAF/MAPK activation, we projected the TERTp mutation status onto a map of cancer cell line cellular states (30). The results are displayed in Fig. 2 where we can see that the TERTp mutation–positive cell lines (Fig. 1A–C) lie on top of the general area of EMT states.

Figure 2A shows the enrichment of the EMT hallmark signature (38) on the map and helps to delineate the EMT states at the bottom of the map. In Fig. 2B, cancer cell lines with TERTp mutations appear to fall across the full span of EMT cancers, including the core EMT state (C4), the partial EMT states (C2, C5, C7, and C9), and the BRAF/MAPK state (C6). Figure 2C shows the enrichment profile of a TERTp signature (“TERTness”) and how closely it resembles the EMT hallmark signature (Fig. 2A). A similar association was apparent when we projected the profiles of the individual mRNA signatures for the two most frequent TERTp mutations C228T and C250T on the same map (Supplementary Fig. S1). Interestingly, a mutation-specific signature for TERT_pg_5.1295250.G.A mutations (see Supplementary Table S1) is more narrowly concentrated in the BRAF/MAPK component (C6, Supplementary Fig. S1B).

The signature also strongly resembles the gene expression profile generated from comparing passaged BJ fibroblasts overexpressing TERT plus the SV40 T antigen versus SV40-BJ cells lacking TERT expression (Fig. 2D; ref. 40). This TERT-driven rescue profoundly impacted gene expression in these slower-growing mesenchymal cells (ref. 40, see Materials and Methods) indicating that either telomere maintenance by TERT, or another TERT function, has dramatic consequences for the cells.

The association between TERTp mutations and the composite TERT transcriptional signature is quite high (Fig. 2C and D). This is more clearly illustrated in the heatmap in Supplementary Fig. S2 showing correspondence between the different TERT signatures. The TERTp 250 signature appears to have a narrower scope centered on BRAF/MAPK activation (Supplementary Fig. S1B), while all the others have a broader context that encompasses the full spectrum of EMT cancers (Supplementary Fig. S1C; Fig. 2C and D).

To further examine the mesenchymal state in conjunction with TERTp status, we directly compared TERTp status with the gene expression patterns of several epithelial and mesenchymal markers across different tissue types and TERTp mutations. Canonical mesenchymal markers such as SNAI1, SNAI2 (SLUG), ZEB1, ZEB2, TWIST1, TWIST2, N-cadherin (CDH2), and VIMENTIN (VIM) showed statistically significant upregulation in TERTp mutants as opposed to TERTp wild-type cell lines (Fig. 3; Supplementary Fig. S3A–S3C). Conversely, mRNA for epithelial markers such as E-cadherin (CDH1) and grainyhead-like transcription factor 2 (GRHL2; ref. 41) showed significant downregulation in TERTp-mutant lines (Fig. 3, Supplementary Fig. S3A–S3C). Importantly, we observed these EMT traits in cells bearing different TERTp mutations and across most tissue types with these mutations (Fig. 3; Supplementary Fig. S3A–S3C), indicating they are a frequently observed characteristic of TERTp-mutant cancers. The relative prominence of these associations suggests TERTp mutants represent a distinct set of functional states.

Analysis of EMT markers on a tissue-specific basis reveal significant associations with TERTp mutations (Supplementary Fig. S3B and S3C). Loss of E-cadherin is associated with a mesenchymal cellular state and metastasis in cancer (42,43) and significantly alters outcomes in preclinical models (44). To determine whether E-cadherin and other proteins associated with cell-to-cell attachment and cellular identity were altered in TERTp-mutant cancers, we analyzed RPPA data for CCLE lines. These results revealed a strong correlation between E-cadherin mRNA and protein expression (Supplementary Fig. S4A) and that TERTp-mutant cancer types have significantly reduced E-cadherin protein levels (Supplementary Fig. S3C and S4B). We also observe a strong correlation between E-cadherin expression and GRHL2, an epithelial marker that positively regulates E-cadherin expression (refs. 45, 46; Supplementary Fig. S4B). Claudin 7, which is an essential component of tight junctions and focal adhesions (47) and another epithelial marker protein, was also lower in TERTp-mutant lines (Fig. 3; Supplementary Fig. S3B and S3C). In contrast, marker proteins for mesenchymal cells, fibronectin 1 (FN1) and N-cadherin (CDH2), were elevated (Fig. 3; Supplementary Fig. S3B and S3C). Tissue-specific analyses revealed that TERTp-mutant bladder cancers frequently retain epithelial markers such as GRHL2 and fail to display key mesenchymal traits (Supplementary Fig. S3B and S3C). Collectively, these results for TERTp mutants indicate that their adhesive state and surface protein–mediated signaling may be distinct from most cancer types that typically lack these mutations.

To determine whether patient tumor samples also displayed significantly reduced E-cadherin gene expression, we analyzed data from The Cancer Genome Atlas (TCGA). Both melanomas and liver cancers with TERTp mutations displayed significantly lower levels of E-cadherin gene expression in clinical samples (Supplementary Fig. S3D).

The GABPα/GABPβ1 heterodimer is an important driver of mutant TERT promoters in multiple tumor-derived cell types and is essential for GBM tumor formation in mouse xenografts (11, 18, 24). Previous studies identified elevated levels of GAPBβ1 in a subset of cell lines with TERTp mutations. To determine whether TERTp mutations selectively occurred in cells with elevated GABPβ1 levels in most tumors, we analyzed a total of 19 cancer types from the CCLE (31) to assess GABPα/β1 levels. Unexpectedly, the results indicated that across a broad range of tissue types total GABPβ1 transcript levels were not consistently elevated in TERTp-mutant cancers (Fig. 3 and 4; Supplementary Fig. S5A). Previous work in glioblastoma identified that specifically the long isoform, GABPβ1L, regulates TERT (24). We therefore examined individual splice variants of GABPβ1. Short isoforms of GABPβ1 lack exon 9 and instead terminate at a stop codon downstream of exon 8 leading to a C-terminus containing an additional 15 amino acids. Long isoforms of GABPβ1 include exon 9 which encodes an additional 50 amino acids of distinct sequence from GABPβ1S. We analyzed CCLE lines for their levels of the main annotated isoforms of GABPβ1L that contain exon 9 (ENST00000380877.3, ENST00000220429.8, ENST00000543881.1) and those that do not (ENST00000359031.4, ENST00000396464.3, ENST00000429662.2). GABPβ1L isoforms were not elevated in TERTp-mutant lines, although ENST00000380877.7 was nearly significant and two forms of GABPβ1S (ENST00000429662.2 and ENST00000359031.4) were significantly elevated (Fig. 4; Supplementary Fig. S5B). These results suggest that although GABPβ1L is critical to drive mutant TERT promoters in GBM, higher levels of GABPβ1L isoforms in cancer cell lines do not appear to broadly explain the selective incidence of TERTp mutations across these 19 tumor types.

Having established that TERTp mutations mark a subset of cancers that share a distinct expression profile, we considered the potential utility of this mark for redeploying anticancer therapies specifically in this context. To address this possibility, we used The Cancer Therapeutics Response Portal (CTRP; refs. 48, 49) to assess whether the growth of TERTp mutants was selectively inhibited by specific compounds compared to TERTp wt cells. The most selective compound was a BRAF inhibitor, dabrafenib (Fig. 5A). Dabrafenib inhibits the enzymatic activity of V600E/K-mutant BRAF and is FDA approved for use in metastatic melanoma. TERTp mutants are significantly enriched for BRAF mutations, suggesting the overall sensitivity of TERTp-mutant lines may be due to the presence of these mutations. We therefore reassessed the EC₅₀ values with respect to BRAF mutations. This analysis indicated that even in the absence of mutant BRAF, >75% of the TERTp-mutant lines displayed reduced growth at relatively low concentrations (Supplementary Fig. S6).

We also observed that a 4:1 combination of MEK1/2 inhibitor + DOT1L inhibitor (selumetinib + BRDA02303741) was significantly more selective against TERTp mutants (P = 0.00082; Fig. 5B); interestingly, the combination of these two drugs was much more effective than either drug on their own. Two other selumetinib combinations were also moderately effective, with PLX-4032 (8:1, Wilcoxon, P = 0.00082, t test, P = 0.00011) and with JQ-1 (4:1, Wilcoxon, P = 0.00082, t test, P = 0.00011; data not shown). These results from the CTRP confirm the functionality of the shared expression profiles of TERTp mutants, particularly a reliance on BRAF/MEK signaling, and suggest that the shared underlying biology of TERTp mutants may be clinically relevant.

The preceding analysis indicated that BRAF/MAPK activation is a major feature of TERTp-mutant cancers. Previous studies of melanoma showed that TERT expression in TERTp mutants relies on this signaling axis. MEK1 and MEK2 (MEK1/2) are effectors of the RAS–RAF signaling axis and they primarily target ERK1 and ERK2, which are major downstream kinases that localize to chromatin (50). To test the broad reliance of TERTp-mutant cancers on MEK1/2 signaling to drive TERT expression, we inhibited MEK1 and MEK2 in cell lines derived from glioblastoma, liver cancer, bladder cancer, and melanoma. Treatments with low doses of trametinib for 24 hours (51) significantly decreased phosphorylation of ERK1 and ERK2 (Supplementary Fig. S7), and in each of the TERTp-mutant cell lines that we tested, TERT mRNA was also substantially decreased (Fig. 6A). Specifically, these data revealed that inhibition of MEK1/2 downregulates TERT mRNA expression in a broad range of TERTp-mutant cells (Fig. 6A; two-tailed t test assuming heteroscedasticity, P = 4.4 × 10⁻⁴, n = 4 cancer types). These results are consistent with expectations from studies in melanoma and thyroid cancer (25–28) and extend the finding to other TERTp-mutant cancer types. To test whether this effect of MEK1/2 inhibition was specific to mutant TERT promoters, we subjected two telomerase-positive, noncancerous cell types to trametinib treatment. We used induced pluripotent stem cells (iPSC) generated from human fibroblasts and spontaneously immortalized keratinocyte cells (HaCaT). These experiments suggested that the observed effect of MEK1/2 inhibition is not specific to mutant TERT promoter alleles, as both iPSC and HaCaT cells displayed significantly reduced TERT mRNA expression after treatment with trametinib (Fig. 6B).

Consistent with a previous report (21), in multiple cell lines we observed the expected loss of FOS mRNA following MEK1/2 inhibition, but not GABPβ1L (Fig. 6C). This finding indicates that additional regulatory mechanisms beyond GABPβ1L transcriptional regulation operate in response to MEK1/2 signaling to drive TERT mRNA expression.

Diverse mechanisms regulating E-cadherin in cancer frequently involve elevated RAS signaling and transcription factors such as ZEB1, CTBP, and SLUG. We found that treatment of TERTp mutants with MEK1/2 inhibition alleviated CDH1 mRNA repression in several cancer types (Supplementary Fig. S8A and S8B). A time course of CDH1 derepression in the TERTp-mutant bladder cancer cell line SCaBER indicated that this de-repression may occur as soon as 6 hours after drug treatment, in contrast to changes in TERT expression, which were not observed until 24 hours (Supplementary Fig. S8C). These results suggest that MAPK signaling may contribute to shaping the adhesive properties of TERTp-mutant cancers.

The developmental transcription factor SNAI2/SLUG was elevated in a significant majority of TERTp-mutant cell lines. SLUG is a major regulator of mesenchymal cell gene expression impacted by RAS signaling (52). Although SLUG has not previously been reported to regulate telomerase expression, the nearly uniformly elevated levels of SLUG in TERTp-mutant cell lines (P = 3.6 × 10⁻²³) suggested a potential role in TERT expression. To test the importance of SLUG for TERT expression, we generated multiple cell lines deficient for SLUG using lentivirus that encodes a previously validated shRNA targeted to exon 3 (34) and selected for stable expression and SLUG knockdown. We successfully generated five lines from four tissue types (U87MG, DAOY, Mel3429, Mel3616, and MDA-MB-231). At the earliest time point assessable for each line, SLUG protein levels were found to be reduced by between 40% and 90% while TERT mRNA was reduced by 40%–60% (Fig. 7A–C). These results suggest that SLUG levels positively influence TERT mRNA expression in several TERTp-mutant cell lines.

We next tested whether this influence of SLUG on TERT expression may result from directly localizing to the TERT locus. The TERT gene has a SLUG consensus-binding sequence CAGGTG (53) in the proximal promoter and also one each at the 5′ and 3′ ends of intron 1 (Fig. 7D). To test whether SLUG is recruited to these motifs, we performed ChIP in mesenchymal progenitor cells (MPC), where SLUG is predicted to be highly expressed and active. We also tested induced pluripotent stem cells (iPSC) for SLUG binding to TERT. No SLUG binding was observed in iPSCs at TERT. However, MPC displayed strong SLUG recruitment at intron 1 (P = 5.8 × 10⁻⁵), but not at the TERT promoter (Fig. 7E). This intronic localization is consistent with the reported binding profile for SLUG in mice where it displays a 10-fold preference for introns (approximately 45% of all loci bound) over proximal promoters (35). To assess whether SLUG occupancy at TERT in MPC persisted upon differentiation, osteocytes were generated from these MPC cells and subjected to SLUG ChIP. Osteocytes displayed little SLUG occupancy at TERT indicating that this recruitment may be a feature of bone marrow–derived MPCs.

We next tested a broad range of TERTp-mutant tumor-derived cell lines for SLUG binding to the TERT locus. We did not detect binding to the TERT promoter in any cells, and most cells displayed no binding to intron 1 (data not shown), suggesting recruitment to this locus was not a common mechanism by which SLUG influenced TERT expression in cancer cells. However, the melanoma line Mel3429 consistently displayed strong SLUG recruitment near intron 1 (Fig. 7F), but not at other nearby positions in the TERT gene or proximal promoter (data not shown). SLUG has been reported to be a direct target of ERK1/2 (54). Consistent with this, SLUG occupancy at TERT in Mel3429 depended on MEK1/2 pathway signaling, as inhibition with 250 nmol/L trametinib for 24 hours followed by SLUG ChIP (Fig. 7F) showed a significant reduction of SLUG occupancy at TERT (P = 0.04, n = 2). These results are consistent with the hypothesis that SLUG may mediate the effect of MEK1/2 on TERT mRNA expression through direct interactions at the TERT locus in a subset of cell lines.

To achieve telomere maintenance, certain tumor lineages (e.g., glioblastoma, hepatocellular cancer, melanoma) rely on TERTp mutations while other tumor types (e.g., AML, prostate, colon) predominantly or exclusively activate telomerase without these mutations. Our study tested the hypothesis that this selective occurrence of TERTp mutations associates with a particular biological signature. To this end, we derived a TERTp signature by comparing gene and protein expression of TERTp mutant versus TERTp wild-type cell lines. Our results from GSEA analysis of CCLE lines revealed that TERTp mutants do indeed share a set of distinct expression profiles that are likely to impact their pathogenicity. Our data suggest that this signature is not a reflection of TERT overexpression, but encompasses a cancer cell state that is marked by promoter mutations and is shared across tissue types. Moreover, these features are likely to generate distinct adhesion and tumor microenvironmental characteristics (e.g., CD44, E-cadherin, N-cadherin, Claudin 7).

TERTp mutations represent an alternative mechanism by which telomerase is activated in a majority of certain tumor types and has been found to cooccur with other mutations (31). We considered that in addition to tumors harboring TERTp mutations, a larger class enriched with this same transcriptional signature reflects a cancer cellular state for which we propose the term “TERTness.” The utility of this concept is the observation that although 90% of cancers express telomerase, this does not make them equivalent in terms of a TERTness signature. TERTp-mutant cancers may be viewed as harboring more TERTness and can be uniquely identified, in a similar way that the concept of BRCAness has been useful to describe the set of tumors without BRCA1/2 mutations (55) with aberrant homologous recombination (HR). Our findings represent an initial report of common features that characterize TERTp mutants as a class and complement previous studies that identified important differences among TERTp mutants (11, 18, 20, 21). Further efforts aimed at integrating these observations with our current study to delineate biologically and clinically relevant TERTp-mutant variables will be informative.

TERTp mutants display a strong association with mesenchymal gene expression. Many tumors of epithelial origin acquire a subset of mesenchymal traits during oncogenesis. Other cancers, including soft tissue tumors, originate from mesenchymal cell types. That TERTp mutants display this association indicates that they undergo EMT or arise from mesenchymal cellular states. A mesenchymal state has major implications for tumors as it has been linked to persistence of therapy-tolerant cells (56–59). We speculate that the mesenchymal nature of TERTp-mutant cancers may provide some biological plasticity for the cells (60). Under selective pressure from therapeutics, this mesenchymal phenotype may facilitate the emergence of altered cellular states that are therapy tolerant creating opportunities for tumor relapse. Thus, whether the TERTp-mutant signature may predispose these cancers to enhanced therapy resistance is of interest.

Some TERTp wild-type cells displayed overlapping phenotypes with TERTp mutants. One possible interpretation is that for cells in mesenchymal or BRAF-driven cellular states, TERTp mutations provide a selective advantage (thus enriching for their occurrence) but they do not represent the only path to immortalization.

The prominent MAPK/RAS pathway signaling in TERTp mutants suggests that it may cooperate with (61) or drive (62, 63) components of this EMT-like state. Importantly, we observed some mutation-specific characteristics to these signatures (Fig. 2; Supplementary Fig. S1 and S2), suggesting the existence of multiple, related driver mechanisms. Previous work in melanoma (64) suggests that examining the correlation between these mutation-specific signatures with patient outcomes may be informative. Specifically, there is a differential distribution of TERT promoter mutations in primary cutaneous melanomas, which are enriched for C250T/-146 mutations, and these tumors frequently cooccur with BRAF and NRAS mutations (22). RAS pathway inhibition has proven effective in the treatment of some cancers, leading us to speculate that TERTp mutations may serve to stratify patients for treatment with MEK1/2 inhibitors. Supporting this rationale, TERTp mutants, regardless of their KRAS or BRAF mutation status, displayed enhanced sensitivity to dabrafenib as well as to a combination MEK+DOT1L inhibitor. Dabrafenib is FDA approved only for BRAF-mutant melanomas. The basis for this selective approval is that in some cell lines with wild-type BRAF, paradoxical activation of the downstream effectors of RAF, MEK1/2, and ERK1/2, has been observed. This paradoxical activation observed in model systems has guided these restrictions in the clinic. However, to our knowledge, those cell types displaying paradoxical activation in these model systems have not been identified as carrying TERT promoter mutations (65–69). In our dataset, why the growth of most cell lines with wild-type BRAF were sensitive to dabrafenib remains unknown. Of note, 3 of the 14 TERTp-mutant cell lines with wild type BRAF showed little growth inhibition, consistent with a previous report for a glioma line of similar genotype (BTL2176; ref. 21); this heterogeneity among wtBRAF/TERTp–mutant cell lines indicates that undiscovered factors can modulate the response to dabrafenib. Interestingly, it was shown that in the TERTp-mutant breast cancer cell line MDA-MB-231, small molecule–mediated dimerization between BRAF and CRAF (a proposed mechanism of this paradoxical activation) was insufficient to recapitulate the activation (68). These studies, and our analysis of CTRP data, suggest that paradoxical activation mechanisms may be less relevant in a subset of tumors, including TERTp mutants.

A key initial report of TERTp mutations in melanoma speculated that TERTp mutations may be responsive to RAS pathway signaling (3) and subsequent reports have identified such sensitivities (25–28). We observed that TERT expression does rely on MEK1/2, but that this was not unique to the mutation status of the TERT promoter because nontumor derived HaCaT cells and iPSCs displayed similar sensitivities. Our previous work identified that mutant TERT promoters exhibit chromatin features more similar to stem cell TERT promoters than to TERTp wild-type promoters. Thus, our results are consistent with our previous findings (19) and support the relevance of those observations to TERT regulation.

GABPβ1L is critical for TERTp mutant gene expression, but we found no evidence that GABPβ1L mRNA expression levels in cell lines broadly explains the distribution of TERTp mutations in different cancer types. This does not exclude the possibility that during early oncogenesis GABPβ1L may positively select for retention of TERTp mutations. However, in contrast to GABPβ1L, the majority of TERTp-mutant cell lines displayed elevated SNAI2/SLUG expression. SLUG is a transcriptional repressor that recruits LSD1 and nuclear receptor corepressor (NCoR) complex and the DNA-binding scaffold protein CtBP1 in a phosphorylation-dependent manner (53, 70, 71), and may also act as a transcriptional activator. SLUG activity is supported by ERK1/2 signaling in breast cancer through phosphorylation (52, 54). In our study, we observed that SLUG bound to the TERT locus in one melanoma cell line where its occupancy was sensitive to MEK1/2 inhibition, suggesting that SLUG can directly influence TERT expression in a MEK1/2-dependent fashion. SLUG also localized to the TERT locus in mesenchymal progenitor cells, but not in osteocytes terminally differentiated from these cells, suggesting a specific association at TERT for primary cells in a mesenchymal state. In contrast, we did not see SLUG at intron 1 of TERT or in the TERT promoter in most cancer cell lines tested, suggesting that in most cases the mechanism by which it influences TERT expression maybe be indirect. Such two-step regulation has been observed whereby ERK1/2-dependent activation of SLUG promotes vimentin mRNA expression in the absence of binding to the vimentin promoter (54), indicating the involvement of additional factors. SLUG genome occupancy has been mapped in mouse cells (35) and in human keratinocytes (72) but has not been examined in cancer cells with TERTp mutations. Thus, such experiments in TERTp-mutant cancers may identify the SLUG circuitry impacting TERT transcription. Future experiments will be required to determine the effect of SLUG on TERT expression in cancers with wild-type TERT promoters. In ALT cells, SNAI1 promotes telomere maintenance that involves telomeric (TERRA) transcription and TERRA was found to influence expression of mesenchymal genes (73). Future studies examining whether in TERTp mutants the elevated levels of SNAI1 or SNAI2 impact telomeric transcription and mesenchymal gene expression may also be informative.

The link between the signature derived from TERTp mutant cancers and TERT-overexpressing BJ fibroblasts and HMECs is not clear (Figs. 1C and 2D; Supplementary Fig. S2). It is not explained by total TERT mRNA expression levels in TERTp-mutant cancers (Fig. 3) because TERTp wild-type cancers displayed similar levels. One possible explanation is that the impact of telomerase on global gene expression is context dependent. If true, this would imply that gene expression differences between such aged cells rescued with telomerase may resemble gene expression differences between wild-type cancers versus TERTp mutants.

About 10%–15% of cancers maintain telomeres although a nontelomerase, homologous recombination-based mechanism called Alternative Lengthening of Telomeres (ALT; recently reviewed in ref. 74). Interestingly, ALT cells also associate with a mesenchymal cellular state (75, 76), suggesting that a mesenchymal state may necessitate alternative pathways to telomere maintenance. Importantly, it has been documented that cancer cells can switch telomere maintenance mechanisms between telomerase and ALT and that these two mechanisms in some cases coexist (77). The TERT gene in ALT cancers is transcriptionally repressed and associates with H3K9me3, H3K27me3, and DNA methylation (78). These features are very similar to the transcriptionally silent TERT allele in cancers with heterozygous TERT promoter mutations (11, 19) suggesting common mechanisms may operate at silent TERT promoters in these two cancer types.

Telomere maintenance is a critical requirement for the indefinite proliferation of cancer cells and underpins their ability to cause fatal disease. Activating mutations in the TERT promoter are the most common recurrent noncoding genetic alterations in cancer (31, 79). Our study reports an unanticipated association between TERTp mutations and a biological state shared across multiple tumor types. Our results suggest that these mutations commonly arise in a specific cellular milieu and may capitalize on this state. This study represents the first global assessment of the biology of TERTp-mutant cancers and suggests that a TERTp-mutant “signature”, or “TERTness,” may exist with implications for a clinical understanding of cancers that share this common mutation.

T.R. Cech is a member of the Board of Directors for Merck and Co., Inc. and a consultant/advisory board member for Storm Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.L. Stern, T.R. Cech, F.W. Huang

Development of methodology: J.L. Stern, G. Hibshman, P. Tamayo, F.W. Huang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.L. Stern, G. Hibshman, T.R. Cech, F.W. Huang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.L. Stern, K. Hu, S.E. Ferrara, J.C. Costello, P. Tamayo, T.R. Cech, F.W. Huang

Writing, review, and/or revision of the manuscript: J.L. Stern, K. Hu, J.C. Costello, W. Kim, P. Tamayo, T.R. Cech, F.W. Huang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.L. Stern, F.W. Huang

Study supervision: J.L. Stern, T.R. Cech, F.W. Huang

We thank T. Rowland, K. Foster, and other Cech lab members, H. Ma, J. Silver, V. Rao, and K. Anseth (University of Colorado, Boulder, CO), K. Couts (University of Colorado, Anschutz Medical Campus, Aurora, CO), B. Lasseigne (University of Alabama at Birmingham, Birmingham, AL). J.L. Stern was supported by a postdoctoral fellowship 127621-PF-16-099-01-DMC from the American Cancer Society and a Mary Ann Harvard Young Investigator Grant from the O'Neal Comprehensive Cancer Center. We thank Y. Zwang for the provision of cell lines. F.W. Huang was supported by a Prostate Cancer Foundation Young Investigator Award. T.R. Cech is an investigator of the Howard Hughes Medical Institute. This project was supported by the following NIH grants: R01GM099705 (to T.R. Cech), R01GM074024 (to P. Tamayo), U24CA194107 (to P. Tamayo), R01HG009285 (to P. Tamayo), U54CA209891 (to P. Tamayo), R01CA172513 (to P. Tamayo), U24CA220341 (to P. Tamayo), U01CA217885 (to P. Tamayo), 5P20CA233255 (to F.W. Huang), and U19CA214253 (to F.W. Huang). Support was received from the Biostatistics and Bioinformatics Shared Resource through the University of Colorado Cancer Center Support Grant (P30CA046934).

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.